{"entities":[{"id":"rsp-2023","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v1.0","publisherId":"anthropic","publishedAt":"2023-09-19","effectiveAt":"2023-09-19","dateNote":"Initial public policy dated September 19, 2023; original entity and publication claim retained."}},{"id":"fsf-2024","type":"document","documentVersion":{"family":"deepmind-fsf","label":"v1.0","publisherId":"google-deepmind","publishedAt":"2024-05-17","dateNote":"The sourced publication date does not establish an implementation or effective date."}},{"id":"o1-card-2024","shortName":"o1 system card · 2024","type":"document","sourceIds":["source-o1-card","mc-src-o1-card-pdf"]},{"id":"rsp-v2-0","shortName":"RSP · v2.0","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v2.0","publisherId":"anthropic","effectiveAt":"2024-10-15","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"rsp-v2-1","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v2.1","publisherId":"anthropic","effectiveAt":"2025-03-31","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"rsp-v2-2","shortName":"RSP · v2.2","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v2.2","publisherId":"anthropic","effectiveAt":"2025-05-14","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"rsp-v3-0","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v3.0","publisherId":"anthropic","publishedAt":"2026-02-24","effectiveAt":"2026-02-24","dateNote":"The cover effective date and release announcement publication date both identify February 24, 2026."}},{"id":"rsp-v3-1","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v3.1","publisherId":"anthropic","effectiveAt":"2026-04-02","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"rsp-v3-2","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v3.2","publisherId":"anthropic","effectiveAt":"2026-04-29","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"rsp-v3-3","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v3.3","publisherId":"anthropic","effectiveAt":"2026-05-26","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"rsp-v3-4","type":"document","documentVersion":{"family":"anthropic-rsp","label":"v3.4","publisherId":"anthropic","effectiveAt":"2026-07-08","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"fsf-v2-0","type":"document","documentVersion":{"family":"deepmind-fsf","label":"v2.0","publisherId":"google-deepmind","publishedAt":"2025-02-04","dateNote":"The sourced publication date does not establish an implementation or effective date."}},{"id":"fsf-v3-0","type":"document","documentVersion":{"family":"deepmind-fsf","label":"v3.0","publisherId":"google-deepmind","publishedAt":"2025-09-22","dateNote":"The sourced publication date does not establish an implementation or effective date."}},{"id":"fsf-v3-1","shortName":"FSF · v3.1","type":"document","documentVersion":{"family":"deepmind-fsf","label":"v3.1","publisherId":"google-deepmind","publishedAt":"2026-04-17","dateNote":"The sourced publication date does not establish an implementation or effective date."}},{"id":"preparedness-beta-2023","shortName":"Preparedness · Beta","type":"document","documentVersion":{"family":"openai-preparedness","label":"Beta","publisherId":"openai","publishedAt":"2023-12-18","dateNote":"The sourced publication date does not establish an implementation or effective date."}},{"id":"preparedness-v2","shortName":"Preparedness · Version 2","type":"document","documentVersion":{"family":"openai-preparedness","label":"Version 2","publisherId":"openai","publishedAt":"2025-04-15","revisedAt":"2025-04-15","dateNote":"Cover last-updated date and dated publication announcement both identify April 15, 2025; no distinct effective date is asserted."}},{"id":"faif-2025","type":"document","documentVersion":{"family":"xai-faif","label":"December 2025 edition","publisherId":"xai","revisedAt":"2025-12-30","dateNote":"The printed update date is retained; it is not silently treated as publication. The URL filename says December 31; the visible document says last updated December 30."}},{"id":"faif-2026","shortName":"FAIF · June 2026 edition","type":"document","documentVersion":{"family":"xai-faif","label":"June 2026 edition","publisherId":"xai","effectiveAt":"2026-06-30","dateNote":"The stated effective date is retained; a separate publication day is not inferred."}},{"id":"claude2-model-card","name":"Claude 2 model card","type":"document","sourceIds":["source-eval-i-claude2-card","mc-src-claude-2-1-card"]},{"id":"claude-3-5-haiku-sonnet-addendum-2024","name":"Model Card Addendum: Claude 3.5 Haiku and Upgraded Claude 3.5 Sonnet","type":"document","sourceIds":["mc-src-claude-3-5-addendum"]},{"id":"claude-opus-4-5-system-card","name":"System Card: Claude Opus 4.5","type":"document","sourceIds":["mc-src-claude-opus-4-5-card"],"documentVersion":{"family":"anthropic-claude-opus-4-5-system-card","label":"As dated November 2025 on its cover, with changelog entries to December 5, 2025","publisherId":"anthropic","publishedAt":"2025-11","revisedAt":"2025-12-05","dateNote":"The cover states 'November 2025', month precision. The changelog lists revisions dated November 24, November 25 and December 5, 2025; its first entry refers to a previous version. The expression read includes the revision of December 5, 2025."}},{"id":"claude-haiku-5-5-system-card","name":"System Card: Claude Haiku 5.5","type":"document","sourceIds":["mc-src-claude-haiku-5-5-card"],"documentVersion":{"family":"anthropic-claude-haiku-5-5-system-card","label":"As dated October 7, 2026 on its cover","publisherId":"anthropic","publishedAt":"2026-10-07","dateNote":"The date printed on the cover, day precision; the card does not say whether it is an issue or publication date. No changelog or revision date was found."}},{"id":"gemini-3-flash-model-card","name":"Gemini 3 Flash Model Card","type":"document","sourceIds":["mc-src-gemini-3-flash-card"],"documentVersion":{"family":"google-deepmind-gemini-3-flash-model-card","label":"As dated 'Published: December 2025' in its preamble","publisherId":"google-deepmind","publishedAt":"2025-12","dateNote":"The card states 'Published: December 2025', month precision; no revision date is printed."}},{"id":"gemini-4-argon-model-card","name":"Gemini 4 Argon — Model Card","type":"document","sourceIds":["mc-src-gemini-4-argon-card"],"documentVersion":{"family":"google-deepmind-gemini-4-argon-model-card","label":"As dated 'Published: October 2026' in its preamble","publisherId":"google-deepmind","publishedAt":"2026-10","dateNote":"The card states 'Published: October 2026', month precision; no revision date is printed."}},{"id":"llama-2-model-card","name":"Llama 2 model card (MODEL_CARD.md, meta-llama/llama)","type":"document","sourceIds":["mc-src-llama-2-card"]},{"id":"llama-3-3-model-card","name":"Llama 3.3 model card (MODEL_CARD.md, meta-llama/llama-models)","type":"document","sourceIds":["mc-src-llama-3-3-card"]},{"id":"cwm-preparedness-report","name":"Code World Model Preparedness Report","type":"document","sourceIds":["mc-src-cwm-preparedness-report"],"documentVersion":{"family":"meta-cwm-preparedness-report","label":"The expression dated October 8, 2025","publisherId":"meta","revisedAt":"2025-10-08","dateNote":"The report prints 'Date: October 8, 2025' without saying whether it is a first publication or a revision. Meta's publication page for the report bears an earlier date; that page is not a source of this record, so no first publication date is recorded."}},{"id":"muse-spark-1-3-evaluation-report","name":"Muse Spark 1.3 Evaluation Report","type":"document","sourceIds":["mc-src-muse-spark-1-3-report"],"documentVersion":{"family":"meta-muse-spark-1-3-evaluation-report","label":"The expression dated October 23, 2026","publisherId":"meta","revisedAt":"2026-10-23","dateNote":"The report prints 'Date: October 23, 2026' without saying whether it is a first publication or a revision. An earlier announcement by Meta links a report at the same address; it is not a source of this record, so no first publication date is recorded."}},{"id":"phi-2-model-card","name":"Phi-2 model card (README.md, microsoft/phi-2)","type":"document","sourceIds":["mc-src-phi-2-card"]},{"id":"phi-4-model-card","name":"Phi-4 model card (README.md, microsoft/phi-4)","type":"document","sourceIds":["mc-src-phi-4-card"]},{"id":"phi-4-mini-flash-reasoning-model-card","name":"Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning)","type":"document","sourceIds":["mc-src-phi-4-mini-flash-reasoning-card"]},{"id":"mai-cyber-1-1-flash-model-card","name":"MAI-Cyber-1.1-Flash model card","type":"document","sourceIds":["mc-src-mai-cyber-1-1-flash-card"]},{"id":"gpt-4v-system-card","name":"GPT-4V(ision) System Card","type":"document","sourceIds":["mc-src-gpt-4v-card"],"documentVersion":{"family":"openai-gpt-4v-system-card","label":"As dated September 25, 2023 on its cover","publisherId":"openai","publishedAt":"2023-09-25","dateNote":"The date printed beneath the title and author line, day precision; the card does not label it as an issue or publication date. No revision date was located."}},{"id":"gpt-5-2-codex-system-card-addendum","name":"Addendum to GPT-5.2 System Card: GPT-5.2-Codex","type":"document","sourceIds":["mc-src-gpt-5-2-codex-addendum"],"documentVersion":{"family":"openai-gpt-5-2-codex-system-card-addendum","label":"As dated December 18, 2025 on its cover","publisherId":"openai","publishedAt":"2025-12-18","dateNote":"The date printed on the cover, day precision; the card does not label its kind. No revision date was located."}},{"id":"gpt-6-luna-october-29-addendum","name":"Addendum to GPT-6 Luna System Card: October 29, 2026 Update","type":"document","sourceIds":["mc-src-gpt-6-luna-october-29-addendum"],"documentVersion":{"family":"openai-gpt-6-luna-october-29-addendum","label":"As dated 2026-10-29 on its cover","publisherId":"openai","publishedAt":"2026-10-29","dateNote":"The date printed on the cover, day precision; the card does not label its kind. The file's address reads 'october-28-2026-addendum'; the difference is recorded, not resolved."}},{"id":"grok-4-1-model-card","name":"Grok 4.1 Model Card","type":"document","sourceIds":["mc-src-grok-4-1-card"],"documentVersion":{"family":"xai-grok-4-1-model-card","label":"As dated November 17, 2025 under its title","publisherId":"xai","publishedAt":"2025-11-17","dateNote":"The date printed under the title, day precision, without a label; the file name carries the same date. No revision date is printed."}},{"id":"grok-4-7-model-card","name":"Model Card: Grok 4.7","type":"document","sourceIds":["mc-src-grok-4-7-card"],"documentVersion":{"family":"xai-grok-4-7-model-card","label":"As dated September 21, 2026 on its cover (Revision: 2026-09-21)","publisherId":"xai","publishedAt":"2026-09-21","revisedAt":"2026-09-21","dateNote":"The cover prints the card's date, without a label, and 'Revision: 2026-09-21'; both are the same day."}}],"claims":[{"id":"ab-ev-o1-card-cb-medium-result","subject":"o1-card-2024","predicate":"states_evaluation_result","object":{"kind":"literal","value":"Medium risk for chemical and biological threat creation: the card states its evaluations found o1 can help experts with the operational planning of reproducing a known biological threat, which meets OpenAI's medium risk threshold."},"topic":"governance","time":{"kind":"event","date":"2024-12-05","basis":"publication","note":"The classification as stated in the card expression of 5 December 2024. The expression checked is the PDF served at cdn.openai.com/o1-system-card-20241205.pdf on 2026-10-28, which carries a footnote in §3 ('Scope of testing') reading 'Section added after December 5th on 12/19/2024'; the served file is therefore a later expression than the card's stated date of 5 December 2024, the card does not say what else differs, and the record keeps the stated date."},"attributes":{"evaluatedSystem":"o1, pre-mitigation and post-mitigation versions as described in the card","evaluation":{"what":"Chemical and biological threat creation capability, against OpenAI's own risk thresholds","how":"A suite of internal evaluations, with risk levels assigned by indicators and reviewed by the Safety Advisory Group","result":"Score for o1: Medium, the same as o1-preview; the Safety Advisory Group classified the pre-mitigation model as overall medium risk, including medium risk for CBRN, and rated post-mitigation risk the same","caveats":"The card states that the exact performance numbers for the model used in production may vary with final parameters, system prompt and other factors, and that the model tested was a near-final post-mitigation model."}}},{"id":"ab-ev-o1-card-non-expert-statement","subject":"o1-card-2024","predicate":"states_evaluation_result","object":{"kind":"literal","value":"The card states that the models do not enable non-experts to create biological threats, because creating such a threat requires hands-on laboratory skills that the models cannot replace."},"topic":"governance","time":{"kind":"event","date":"2024-12-05","basis":"publication","note":"As stated in the card expression of 5 December 2024. The expression checked is the PDF served at cdn.openai.com/o1-system-card-20241205.pdf on 2026-10-28, which carries a footnote in §3 ('Scope of testing') reading 'Section added after December 5th on 12/19/2024'; the served file is therefore a later expression than the card's stated date of 5 December 2024, the card does not say what else differs, and the record keeps the stated date."},"attributes":{"evaluatedSystem":"o1, as evaluated for the December 2024 card","evaluation":{"what":"Whether the evaluated models enable non-experts to create biological threats","how":"The card's reading of its own chemical and biological evaluation suite","result":"The models do not enable non-experts, because the card states that creating such a threat requires hands-on laboratory skills the models cannot replace","caveats":"Stated as part of the same summary that classifies the model as medium risk for expert operational planning; the card gives no measurement for the non-expert case."}}},{"id":"mc-claude-3-5-addendum-cites-rsp-v2-0","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"cites","object":{"kind":"entity","entityId":"rsp-v2-0"},"topic":"governance","time":{"kind":"unknown","note":"The citing card prints no date of its own."}},{"id":"mc-claude-opus-4-5-card-cites-rsp-v2-2","subject":"claude-opus-4-5-system-card","predicate":"cites","object":{"kind":"entity","entityId":"rsp-v2-2"},"topic":"governance","time":{"kind":"event","date":"2025-11","basis":"publication","note":"Dated as the citing card is dated; not the date of the cited edition."}},{"id":"mc-gemini-4-argon-card-cites-fsf-v3-1","subject":"gemini-4-argon-model-card","predicate":"cites","object":{"kind":"entity","entityId":"fsf-v3-1"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the citing card is dated; not the date of the cited edition."}},{"id":"mc-o1-card-cites-preparedness-beta","subject":"o1-card-2024","predicate":"cites","object":{"kind":"entity","entityId":"preparedness-beta-2023"},"topic":"governance","time":{"kind":"event","date":"2024-12-05","note":"Dated as the citing card is dated; not the date of the cited edition."}},{"id":"mc-gpt-5-2-codex-addendum-cites-preparedness-v2","subject":"gpt-5-2-codex-system-card-addendum","predicate":"cites","object":{"kind":"entity","entityId":"preparedness-v2"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the citing card is dated; not the date of the cited edition."}},{"id":"mc-gpt-6-luna-october-29-addendum-cites-preparedness-v2","subject":"gpt-6-luna-october-29-addendum","predicate":"cites","object":{"kind":"entity","entityId":"preparedness-v2"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the citing card is dated; not the date of the cited edition."}},{"id":"mc-grok-4-7-card-cites-faif-2026","subject":"grok-4-7-model-card","predicate":"cites","object":{"kind":"entity","entityId":"faif-2026"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the citing card is dated; not the date of the cited edition."}},{"id":"mc-cell-anthropic-claude-2-1-q01","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states a revision date (November 21, 2023, day precision) for the addition of the Claude 2.1 appendix. No issue date of the card is printed; Anthropic's index lists the Claude 2 card under July 2023. Search for the rest of the criterion: Whole text (16 pages), including the title block and the appendix title block (PDF p. 15). Terms searched: “2023”, “July”, “November”, “Update”, “date”. Passages read and not counted: References, PDF pp. 12–14: Access dates of cited works (2023-07-03 to 2023-07-08), not the card's date. Section 2 'Model details', PDF p. 1: 'released in July 2023' is the model's release month (Q02). Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-2-1-q02","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states Claude 2's release month (July 2023). For Claude 2.1 the appendix states a new release without a date; the appendix also states that support at Claude 2.0's initial product launch was 100K tokens and is now 200K. The November 21, 2023 date on PDF p. 1 is the card’s update, not a release date. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q02-1","mc-ev-cell-anthropic-claude-2-1-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-2-1-q03","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card names the developer’s framework (Responsible Scaling Policy, Preparedness Framework, Frontier Safety Framework, Risk Management Framework, Frontier AI Framework, Advanced AI Scaling Framework, Frontier Governance Framework, or a successor). Stated if it names the framework and identifies the edition (version number, date or a link to a dated edition); partly if it names the framework without an edition. Not applicable if the card predates every framework edition of its developer. The inspected scope of Claude 2 model card was: Whole text (16 pages), including references and the Claude 2.1 appendix. Terms searched: “Responsible Scaling”, “RSP”, “ASL”, “AI Safety Level”, “scaling policy”, “framework”, “commitment”. Passages read and not counted: Section 2, 'Ethical Considerations', PDF p. 2: The Acceptable Use Policy (AUP) is a usage policy, not a frontier safety framework. The conclusion is bounded by that scope and method.","qualification":"No passage naming a frontier safety framework was found. The card's update line (2023-11-21) postdates the publication the dataset records for RSP v1.0 (2023-09-19), so the question is asked. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q03"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03"}}}},{"id":"mc-cell-anthropic-claude-2-1-q04","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A hyperlink, address or bibliographic reference to the framework document in the card. Partly if it refers to the framework as a document without an address or reference. Not applicable if Q03 is not applicable. The inspected scope of Claude 2 model card was: Whole text, including the reference list (PDF pp. 12–14). Terms searched: “Responsible Scaling”, “RSP”, “policy”, “framework”. Passages read and not counted: References [11], PDF p. 12: Reference to the Acceptable Use Policy, not a frontier safety framework. The conclusion is bounded by that scope and method.","qualification":"No framework is named (Q03), and no link or reference to a framework document was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q04"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04"}}}},{"id":"mc-cell-anthropic-claude-2-1-q05","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A sentence stating the outcome for the model in the framework’s own categories (for example, a safety level applied, a capability threshold reached or not reached, a risk category assigned), quoted. Partly if the card says a determination was made without stating its outcome, or states an outcome for some tracked domains only. Not applicable if Q03 is not applicable. The inspected scope of Claude 2 model card was: Section 2 'Evaluations and Red Teaming' and the whole text. Terms searched: “threshold”, “level”, “ASL”, “determination”, “risk”. Passages read and not counted: Section 2 'Evaluations and Red Teaming', PDF p. 2: The statement that deployed Claude models are not believed to pose national security or significant safety risks is a risk conclusion, not expressed in a framework's categories. The conclusion is bounded by that scope and method.","qualification":"No framework is named (Q03); no outcome stated in a framework's categories was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q05"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05"}}}},{"id":"mc-cell-anthropic-claude-2-1-q06","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that mitigations contribute to its risk conclusion, without naming them there or tying them to a framework determination or threshold (no framework is named, Q03). Elsewhere it states Constitution-based training, AUP enforcement and Trust and Safety processes. Search for the rest of the criterion: Section 2 (Ethical Considerations, Evaluations and Red Teaming), Section 3.2. Terms searched: “mitigation”, “safeguard”, “Trust and Safety”, “AUP”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-2-1-q07","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of Claude 2 model card was: Whole text. Terms searched: “biolog”, “chemical”, “CBRN”, “weapon”, “national security”, “dangerous capabilit”. Passages read and not counted: Section 2 'Evaluations and Red Teaming', PDF p. 2: 'We evaluated and red-teamed Claude 2 and previous Claude models for several national security and safety related risks' names no domain. The conclusion is bounded by that scope and method.","qualification":"No passage reporting chemical or biological evaluations was found. The card states evaluation and red-teaming for national security risks without naming domains. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-anthropic-claude-2-1-q08","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of offensive cyber capability were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. The inspected scope of Claude 2 model card was: Whole text. Terms searched: “cyber”, “hack”, “exploit”, “vulnerab”, “security”, “national security”. Passages read and not counted: Section 2 'Evaluations and Red Teaming', PDF p. 2: National security evaluations are stated without naming a domain. Section 3.4, PDF pp. 5–6: 'Jailbreaks' are attempts to elicit harmful output, not cyber-offence evaluations. The conclusion is bounded by that scope and method.","qualification":"No passage reporting offensive cyber evaluations was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q08"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08"}}}},{"id":"mc-cell-anthropic-claude-2-1-q09","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that ARC's safety audits addressed autonomous replication and that neither ARC nor Anthropic believes current Claude models have that capability. This is Anthropic's account; ARC's own report is not part of the card. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q09-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-2-1-q10","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Claude 2 model card was: Whole text. Terms searched: “persua”, “manipulat”, “influence”, “propaganda”, “misinformation”. Passages read and not counted: Section 2 'Evaluations and Red Teaming', PDF p. 2: Misinformation is named as a Trust and Safety area for future red-teaming, not an evaluation of persuasive capability. The conclusion is bounded by that scope and method.","qualification":"No passage reporting evaluations of persuasion or manipulation of people was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-anthropic-claude-2-1-q11","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Claude 2 model card was: Section 3 'Alignment Evaluations' (3.1–3.5) and the appendix 'Honesty Evaluations'. Terms searched: “decept”, “deceiv”, “scheming”, “sabotage”, “reward hack”, “oversight”, “honesty”, “alignment”. Passages read and not counted: Section 3 'Alignment Evaluations', PDF pp. 3–7: Human preference Elo, BBQ bias, TruthfulQA, harmfulness and HHH evaluations measure helpfulness, harmlessness, bias and truthfulness, not deception or sabotage. Appendix Section 3 'Honesty Evaluations', PDF p. 16: Factual accuracy and faithfulness to documents are accuracy measures, not deception propensity. The conclusion is bounded by that scope and method.","qualification":"No evaluation of the model's own propensity or capability for deception, scheming, reward hacking or sabotage was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-anthropic-claude-2-1-q12","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states pre-deployment red-teaming by external red teamers including crowdworker platforms on Trust and Safety topics, and red-teaming for unnamed national security risks. Section 3.1 adds a crowdworker red-teaming task and Section 3.4 automated red-teaming on 328 held-out prompts. No number of testers or period is given. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q12-1","mc-ev-cell-anthropic-claude-2-1-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-2-1-q13","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card names ARC as conducting safety audits of Anthropic's models; the crowdworker platforms used for red-teaming are not named. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-2-1-q14","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Claude 2 model card was: Whole text. Terms searched: “government”, “policymakers”, “institute”, “AISI”, “public sector”, “national security”. Passages read and not counted: Section 2 'Evaluations and Red Teaming', PDF p. 2: 'We are working with policymakers and other labs to share our findings' concerns sharing findings, not testing or access. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested the model or received access. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-anthropic-claude-2-1-q15","subject":"claude2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Claude 2 model card was: Sections 1, 2 and 5, and the appendix introduction. Terms searched: “decision”, “decide”, “deploy”, “release”, “launch”, “approve”. Passages read and not counted: Section 2 'Model details', PDF p. 1: 'released in July 2023' states the release as a fact, not a decision. Section 2 'Evaluations and Red Teaming', PDF p. 2: The belief that deployed Claude models do not pose national security risks is a risk conclusion, not a release decision. Section 4.2 'Long Contexts', PDF p. 9: 'we will support 100K at launch rather than this full context window' concerns a product setting at launch, not the decision to release. The conclusion is bounded by that scope and method.","qualification":"No passage stating that a decision to release or deploy was made was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-anthropic-claude-2-1-q16","subject":"claude2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2023-11-21","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states a limitation of its crowdworker-based human feedback and red-teaming evaluation, and that the Claude 2.1 document-faithfulness evaluation is at an early stage. It also states that it expects to release new findings as its research and evaluations continue (Section 1). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-2-1-q16-1","mc-ev-cell-anthropic-claude-2-1-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q01","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of Model Card Addendum: Claude 3.5 Haiku and Upgraded Claude 3.5 Sonnet was: Title block (PDF p. 1) and whole text, including footnotes and references. Terms searched: “2024”, “October”, “date”, “update”, “version”. Passages read and not counted: Section 2.2, footnote 2, PDF p. 3: 'as of October 22nd, 2024' dates a cited leaderboard reading, not the card. References [4], PDF p. 13: 'October, 2024' dates the cited Responsible Scaling Policy, not the card. The conclusion is bounded by that scope and method.","qualification":"No date of the card was found in the document. Anthropic's index lists it under October 2024. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q02","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states a release form for Claude 3.5 Haiku (initial launch as text-only) and refers to the upgraded Claude 3.5 Sonnet's release, without a release or availability date for either model. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q02-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q03","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card names the Responsible Scaling Policy and identifies the edition by a dated reference (October 2024) whose address carries the date 2024-10-15, which the dataset records for Responsible Scaling Policy v2.0. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q03-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q03-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q04","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card gives a bibliographic reference with the address of the policy document. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q05","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states for the upgraded Claude 3.5 Sonnet that ASL-3 safeguards are not required, and for both models that they did not trigger the policy's full evaluation protocol. It also states that the computer use capability is classified under the ASL-2 framework (Section 2.1). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q05-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q05-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q06","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states safeguards, monitoring and new classifiers, but does not tie them to the ASL determination in Q05 or name a safeguard standard applied; the determination states that ASL-3 safeguards are not required. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q06-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q07","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states CBRN evaluations under the RSP and, for the upgraded Claude 3.5 Sonnet, that no domain showed capabilities requiring ASL-3 safeguards and security. Radiological and nuclear risks are reported together with chemical and biological as CBRN. No domain outcome is stated separately for Claude 3.5 Haiku. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q07-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q07-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q08","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states cybersecurity evaluations under the RSP and, for the upgraded Claude 3.5 Sonnet, that no domain showed capabilities requiring ASL-3 safeguards and security. Method descriptions in Section 3.2 are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q08-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q08-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q09","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states autonomy evaluations under the RSP, that its software engineering evaluations suffice to rule out ASL-3 autonomy capabilities, and that the developer judges the upgraded Claude 3.5 Sonnet not to require ASL-3 safeguards. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q09-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q09-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q10","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Model Card Addendum: Claude 3.5 Haiku and Upgraded Claude 3.5 Sonnet was: Sections 1–3 (whole text). Terms searched: “persua”, “manipulat”, “influence”, “election”, “propaganda”. Passages read and not counted: Section 3.1.1 'Model Red-Teaming', PDF p. 7: 'Elections Integrity' is a Trust & Safety policy area for harmful requests, not an evaluation of persuasive capability. The conclusion is bounded by that scope and method.","qualification":"No passage reporting evaluations of persuasion or manipulation of people was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q11","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Model Card Addendum: Claude 3.5 Haiku and Upgraded Claude 3.5 Sonnet was: Sections 2.4–2.5 and 3 (whole text). Terms searched: “decept”, “deceiv”, “scheming”, “sabotage”, “reward hack”, “sandbag”, “oversight”, “honesty”, “alignment”. Passages read and not counted: Section 3.1.1, Figure 2, PDF p. 8: 'Honesty' human preference win rates are preference ratings, not an evaluation of deception or scheming. Section 3.1.2 'Prompt Injection', PDF p. 8: Robustness to third-party attacks, not the model's own propensity. Section 1 'Introduction', PDF p. 1: 'alignment with Anthropic’s Usage Policy' refers to usage-policy compliance. The conclusion is bounded by that scope and method.","qualification":"No evaluation of the models' own propensity or capability for deception, scheming, reward hacking or sabotage was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q12","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states Trust & Safety red-teaming across fourteen policy areas in six languages, computer-use red-teaming, and first- and third-party multimodal red-teaming; Section 3.2 also mentions manual red-teaming within CBRN evaluations. No number of testers or period is given. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q12-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q12-2","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q12-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q13","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card names METR as conducting an independent assessment; it does not state the assessment's scope or outcome, or which model it covered. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q14","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states pre-deployment testing of the upgraded Claude 3.5 Sonnet by the US and UK AI Safety Institutes; no government testing of Claude 3.5 Haiku is stated. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q14-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q15","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states a safeguards determination in the company's voice and refers to the release, but no passage states that a decision to release was made or by whom. Search for the rest of the criterion: Sections 1, 2.1, 3.1.3, 3.2 and 3.3. Terms searched: “decision”, “decide”, “deploy”, “release”, “launch”, “Responsible Scaling Officer”, “CEO”. Passages read and not counted: Section 3.1.3, PDF p. 9: 'a safe and measured release of this capability' concerns monitoring of the computer use feature, not a release decision. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q15-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q15-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q16","subject":"claude-3-5-haiku-sonnet-addendum-2024","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states that a reported agentic coding result could be improved with dedicated scaffolding and prompting, and describes its reasoning on computer use and RSP threat models as preliminary. No limitation is stated for the Section 3.2 frontier risk results. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q16-1","mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q01","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card's printed date has month precision (November 2025). Its changelog lists three revision dates (November 24 and 25, December 5, 2025); the first entry refers to a 'previous version' of the card. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q01-1","mc-ev-cell-anthropic-claude-opus-4-5-q01-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q02","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that the model has been deployed and refers to its launch, naming Claude.ai as a surface, but gives no release or availability date. Search for the rest of the criterion: Abstract, Sections 1–1.2, and term search across the card. Terms searched: “release date”, “available”, “availability”, “launch”, “API”, “Claude.ai”, “November 2”. Passages read and not counted: Section 1.2.4, PDF pp. 12–13: 'We are launching Claude Opus 4.5 with safeguards we believe are appropriate' gives relative timing; the sentence spans a page break in the text layer and is not quoted. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q02-1","mc-ev-cell-anthropic-claude-opus-4-5-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q03","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card names the Responsible Scaling Policy; the text gives no version, but the PDF's hyperlink on these words points to a policy PDF whose cover reads 'Version 2.2, Effective May 14, 2025' (Responsible Scaling Policy v2.2; identified by link annotation, maintainer decision 4a). The printed text does not name the edition. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q04","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card refers the reader to the policy; the words carry a hyperlink to the RSP v2.2 PDF (also on PDF p. 117). The address is not printed in the text. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q05","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that the model does not cross the AI R&D-4 or CBRN-4 thresholds and that it is deployed under the ASL-3 Standard. For cyber it states that the policy sets no capability threshold (Q08). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q05-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q06","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that ASL-3 protections were implemented on the basis of the capability assessment. This is the developer's account of the safeguards applied. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q06-1","mc-ev-cell-anthropic-claude-opus-4-5-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q07","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states biological evaluations with the outcome that the CBRN-4 threshold is not crossed and ASL-3 safeguards are appropriate. It states that chemical risks are not evaluated internally, and that radiological and nuclear evaluations are carried out with NNSA and their results are not published by Anthropic. Evaluation details are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q07-1","mc-ev-cell-anthropic-claude-opus-4-5-q07-2","mc-ev-cell-anthropic-claude-opus-4-5-q07-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q08","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states cyber evaluations and a conclusion without a policy threshold, which it states the policy does not set for cyber. Evaluation methods and scores are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q08-1","mc-ev-cell-anthropic-claude-opus-4-5-q08-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q09","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states AI R&D evaluations, that the AI R&D-4 threshold is not crossed, and that the rule-out drew also on an internal survey and qualitative impressions because benchmark rule-out thresholds were roughly reached. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q09-1","mc-ev-cell-anthropic-claude-opus-4-5-q09-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q10","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card lists attempts to manipulate users toward certain views among the behaviours its alignment assessment tested for; no outcome specific to manipulation or persuasion was found. The overall alignment conclusions are recorded in Q11. Search for the rest of the criterion: Sections 3.5, 6.1–6.2 and term search across the card. Terms searched: “persua”, “manipulat”, “influence”, “sway”, “political”. Passages read and not counted: Section 3.5.1 'Political bias', PDF pp. 43–45: Even-handedness evaluation measures bias in responses, not persuasive or manipulative capability. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q10-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q11","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states alignment evaluations covering sabotage, deception, sandbagging, reward hacking and evaluation awareness, with the conclusion that high-stakes sabotage risk is low. It also reports two isolated cases of deception by omission in testing earlier snapshots, which it attributes to interventions against prompt injection. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q11-1","mc-ev-cell-anthropic-claude-opus-4-5-q11-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q12","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states biological red-teaming by third parties, by one external expert over two days (Section 7.2.4.9) and by CAISI over a one-week window; prompt-injection red-teaming with Gray Swan (Section 5.2); and that cyber capabilities are red-teamed (Section 7.4). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q12-1","mc-ev-cell-anthropic-claude-opus-4-5-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q13","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card names Gray Swan, which it calls an external research partner, as measuring prompt-injection attack success. The third parties in biological red-teaming and the external expert are not named; the changelog refers to ARC-AGI scores 'reported by the ARC Foundation'. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q14","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states pre-deployment testing by CAISI and the UK AISI (CBRN, cyber, ASL-3 safeguards and misalignment), UK AISI testing of a pre-release snapshot for misalignment (Section 6.13), and an NNSA partnership evaluating models for nuclear and radiological risks. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q14-1","mc-ev-cell-anthropic-claude-opus-4-5-q14-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q15","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states a decision to release in the company's voice, and names the Responsible Scaling Officer and CEO as making the ASL determination on which the release rests. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q15-1","mc-ev-cell-anthropic-claude-opus-4-5-q15-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-opus-4-5-q16","subject":"claude-opus-4-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2025-12-05","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that ruling out the AI R&D-4 and CBRN-4 thresholds is becoming difficult and that parts of the thresholds carry epistemic uncertainty. It also states limited tools for studying reasoning faithfulness (Section 6.1.1). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-opus-4-5-q16-1","mc-ev-cell-anthropic-claude-opus-4-5-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q01","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card carries a printed date with day precision (October 7, 2026); its kind (issue or publication) is not stated. No changelog or revision date was found. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q02","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states the release form (general access; Section 1.5 also refers to first-party products and the API) but no release or availability date. Search for the rest of the criterion: Executive summary, Section 1 and term search across the card. Terms searched: “release”, “available”, “availability”, “launch”, “general access”, “October”. Passages read and not counted: Section 8.9.1, PDF p. 125: 'as of October 7, 2026' dates a price used for a comparison model, not the release. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q02-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q03","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the Responsible Scaling Policy without a version or date; its hyperlink points to the undated policy page. As context, not as the card's statement, the dataset records Responsible Scaling Policy v3.4 (effective 2026-07-08) as the edition published before the card's date. The card also names a 'Frontier Compliance Framework (FCF)', a compliance document not yet read and not recorded as a framework edition (maintainer decision 5a). Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q04","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The words 'Responsible Scaling Policy (RSP)' carry a hyperlink to https://www.anthropic.com/responsible-scaling-policy, an undated page; the address is not printed in the text. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q05","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that the model does not cross the CB-2 or Autonomy-2 thresholds and is treated as meeting CB-1 and Autonomy-1. No RSP threshold is named for cyber. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q05-1","mc-ev-cell-anthropic-claude-haiku-5-5-q05-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q06","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that the mitigations corresponding to the CB-1 and Autonomy-1 thresholds are applied, and lists deployed classifiers for CB, cyber, frontier-LLM development and conventional weapons. This is the developer's account of the safeguards. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q06-1","mc-ev-cell-anthropic-claude-haiku-5-5-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q07","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states automated chemical and biological (CB) evaluations, with the outcome that CB-2 is not crossed and CB-1 is treated as met. No radiological or nuclear evaluation is reported in the RSP section. Evaluation names and results are not recorded (hazard rule). Not cross-family checked: the maintainer excluded this cell on chemical and biological evaluations from the cross-family check of 2026-11-03. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q07-1","mc-ev-cell-anthropic-claude-haiku-5-5-q07-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q08","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states cyber capability evaluations and the conclusion of a relative lack of advanced cybersecurity capabilities, compared with other Claude models; no threshold language is used for cyber. Benchmark contents and scores are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q08-1","mc-ev-cell-anthropic-claude-haiku-5-5-q08-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q09","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that the Autonomy-2 threshold is not crossed, carrying over the Claude Opus 5.5 System Card's conclusion on the basis of relative capability, and that Autonomy threat model 1 applies (Q11). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q09-1","mc-ev-cell-anthropic-claude-haiku-5-5-q09-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q10","subject":"claude-haiku-5-5-system-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of System Card: Claude Haiku 5.5 was: Executive summary, Sections 2, 4.1, 6.1–6.2 and term search across the card. Terms searched: “persua”, “manipulat”, “influence”, “propaganda”, “sway”, “convinc”. Passages read and not counted: Section 4.1, PDF p. 30: 'influence operations' is a domain of multi-turn harmful-request testing (content safety), not an evaluation of the model's persuasive capability. Section 6.2 metric list, PDF p. 70: 'Internalized propaganda' is a behavioural-audit metric for promoting politically motivated false claims, not an evaluation of persuasion of people. Section 6.3.3.4, PDF pp. 83–84: 'git manipulations' refers to concealing repository changes (Q11), not manipulation of people. The conclusion is bounded by that scope and method.","qualification":"No passage reporting evaluations of persuasive or manipulative capability was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q11","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states alignment evaluations (behavioural audit, honesty, sandbox escape, covert-capability and sabotage evaluations) with the conclusion that misalignment risk is low. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q11-1","mc-ev-cell-anthropic-claude-haiku-5-5-q11-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q12","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card reports prompt-injection red-teaming under a section titled 'External red teaming' (coding, tool use and computer use surfaces) and adversarial testing by crowd workers (Section 1.2). It states that no expert red-teaming sessions were run for CB (Q07). No number of testers or period is given. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q12-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q13","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names Gray Swan as running its attacker against the latest models. Section 1.6 states that external testers ran assessments without naming them there; Irregular is named as the developer of a cyber benchmark, not as a tester. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q14","subject":"claude-haiku-5-5-system-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of System Card: Claude Haiku 5.5 was: Sections 1.6, 2, 3, 5.2 and term search across the card. Terms searched: “AISI”, “CAISI”, “Center for AI Standards”, “Security Institute”, “government”, “NNSA”, “Department of Energy”. Passages read and not counted: Section 5.2.1 'External red teaming', PDF p. 50: The UK AI Security Institute and the US Center for AI Standards and Innovation are named as partners in creating a benchmark that Anthropic used, not as having tested this model. Section 6 (chain-of-thought controllability), PDF p. 93: The UK AI Security Institute is named as the source of a prompt used by the developer, not as having tested or received access to this model. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested this model or received access to it. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q15","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states, in the company's voice, that the model is being released with a named safeguard set as a consequence of its chemical and biological judgement. It does not describe a decision to release, and names no person, office or committee. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q15-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"partly"}}}},{"id":"mc-cell-anthropic-claude-haiku-5-5-q16","subject":"claude-haiku-5-5-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2026-10-07","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states limits of its alignment measures and of the evidence for this model; it also states that internal testing runs at a smaller scale than deployment (Section 6.1.1) and that CB evaluations were automated, without studies requiring human participants (Q07). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-anthropic-claude-haiku-5-5-q16-1","mc-ev-cell-anthropic-claude-haiku-5-5-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q01","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card carries a publication date at month precision only; the cover repeats it as 'Model card published: December, 2025'. The developer's model-card index lists the card as 'Updated 17 December 2025' (index, not the card). No revision date is printed. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q02","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage of the card giving the model’s release, launch or availability date, or the release form (API, consumer product, open weights, limited preview), quoted. Stated if a date or a dated event is given; partly if only the form of release, or only a relative timing (for example, “at launch”), is given. The inspected scope of Gemini 3 Flash Model Card was: Whole card (6 PDF pages): cover, preamble, Model Information, Model Data, Implementation and Sustainability, Distribution, Evaluation, Intended Usage and Limitations, Ethics and Content Safety. Terms searched: “release”, “launch”, “availab”, “distribut”, “deploy”, “API”, “preview”. Passages read and not counted: Distribution, PDF p. 3 (printed p. 2): States that the model 'is distributed similarly to Gemini 3 Pro' and refers to the Gemini 3 Pro model card; gives neither a date nor a release form. Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5): States the model was deemed acceptable for deployment (recorded under Q15); no date or form of release. Evaluation, 'Results', PDF p. 4 (printed p. 3): 'Results as of December, 2025' dates the evaluation results, not the release. The conclusion is bounded by that scope and method.","qualification":"No release or availability date and no release form was found in the card; its Distribution section refers to the Gemini 3 Pro model card. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q02"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q03","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the Frontier Safety Framework (also 'our FSF') without an edition, date or link to the framework. Context only, not the card's statement: the dataset records Frontier Safety Framework v3.0 (2025-09-22) as the edition published before the card's date. Search for the rest of the criterion: Whole card. Terms searched: “Frontier Safety Framework”, “FSF”, “version”, “edition”, “2025”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q04","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card refers to the framework as a document ('outlined in') without an address or reference. The paragraph's hyperlink (PDF link annotation, p. 6) points to the Gemini 3 Pro Frontier Safety Framework Report (storage.googleapis.com/deepmind-media/gemini/gemini_3_pro_fsf_report.pdf), a model report, not the framework. Search for the rest of the criterion: Whole card and its PDF link annotations (17 links). Terms searched: “Frontier Safety Framework”, “FSF”, “frontier-safety”, “http”. Passages read and not counted: PDF link annotation, p. 6: Target is the Gemini 3 Pro Frontier Safety Framework Report, not the framework document. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q05","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that Gemini 3 Pro Preview did not reach any critical capability level and that Gemini 3 Flash, as a less capable model, is unlikely to reach any CCL; it states that it relies on the results reported for Gemini 3 Pro and reports no frontier safety evaluation of Gemini 3 Flash itself. No domain is named. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q05-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q06","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage naming safeguards, protections or a safeguard standard applied because of, or in line with, the determination or threshold in Q05 (for example, a deployment or security standard named in the framework). Partly if safeguards are stated without being tied to a determination or threshold. A stated safeguard is the developer’s account, not evidence that it works. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “safeguard”, “mitigation”, “protect”, “security”, “filter”, “risk acceptance”. Passages read and not counted: Ethics and Content Safety, 'Risks and Mitigations', PDF p. 6 (printed p. 5): Refers to the Gemini 3 Pro model card; names no safeguard. Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5): Ties the deployment judgement to the FSF's risk acceptance criteria (Q15) but names no safeguard or mitigation. The conclusion is bounded by that scope and method.","qualification":"No safeguard tied to the determination was found; the card refers to the Gemini 3 Pro model card for risks and mitigations. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q06"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q07","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “CBRN”, “chemical”, “biolog”, “radiolog”, “nuclear”, “weapon”, “CCL”, “Frontier Safety”. Passages read and not counted: Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5): Reports another model's (Gemini 3 Pro Preview) results without naming domains; recorded under Q05. The conclusion is bounded by that scope and method.","qualification":"No chemical or biological evaluation of Gemini 3 Flash was found. The card's frontier safety paragraph reports Gemini 3 Pro Preview's results as a whole (no CCL reached) without naming any domain. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q08","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of offensive cyber capability were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “cyber”, “offens”, “exploit”, “vulnerab”, “CCL”. Passages read and not counted: Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5): Reports Gemini 3 Pro Preview's results without naming domains; recorded under Q05. The conclusion is bounded by that scope and method.","qualification":"No cyber evaluation was found; the term does not occur in the card. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q08"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q09","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “autonom”, “R&D”, “machine learning”, “self-repl”, “self-improv”, “agentic”. Passages read and not counted: Evaluation, 'Approach', PDF p. 4 (printed p. 3): Lists 'agentic tool use' among capability benchmarks; product capability, not a dangerous-capability evaluation. Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5): Reports Gemini 3 Pro Preview's results without naming domains; recorded under Q05. The conclusion is bounded by that scope and method.","qualification":"No dangerous-capability evaluation of autonomy, self-improvement or AI R&D was found. The card reports agentic benchmarks as product capability; they are set aside below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q10","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “persua”, “manipul”, “influence”. Passages read and not counted: Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5): Reports Gemini 3 Pro Preview's results without naming domains; recorded under Q05. The conclusion is bounded by that scope and method.","qualification":"No persuasion or manipulation evaluation was found; the terms do not occur in the card. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q11","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “decepti”, “misalign”, “schem”, “sabotag”, “situational”, “stealth”, “oversight”, “reward hack”, “honest”. Passages read and not counted: Ethics and Content Safety, 'Training and Development Evaluation Results', PDF p. 5 (printed p. 4): Content safety, tone and unjustified-refusal evaluations; content-safety tests do not meet the criterion. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The safety evaluations reported are content-safety, tone and refusal measures. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q12","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states manual red teaming by Google specialist teams outside the model development team, covering child safety and content safety policies and issues beyond them; it states no number of testers or period. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q12-1","mc-ev-cell-google-deepmind-gemini-3-flash-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q13","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “external”, “third party”, “third-party”, “partner”, “independent”, “vendor”, “evaluator”. Passages read and not counted: Ethics and Content Safety, 'Human Red Teaming Results', PDF p. 6 (printed p. 5): 'specialist teams who sit outside of the model development team' are Google teams (Q12), not external organisations. The conclusion is bounded by that scope and method.","qualification":"No external tester or evaluator was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q14","subject":"gemini-3-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Gemini 3 Flash Model Card was: Whole card. Terms searched: “institute”, “AISI”, “CAISI”, “government”, “UK”, “U.S.”. The conclusion is bounded by that scope and method.","qualification":"No government body was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q15","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states a deployment judgement in the company's voice ('we'), made 'in line with the risk acceptance criteria outlined in our FSF'. No person, office or committee is named. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q15-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-3-flash-q16","subject":"gemini-3-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2025-12","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that the reported safety results are automated only, are computed with revised evaluations not comparable with earlier cards, and that variation is expected in automated results. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-3-flash-q16-1","mc-ev-cell-google-deepmind-gemini-3-flash-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q01","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The PDF carries a publication date at month precision only and no revision date. The card's HTML page reads 'Published 13 October 2026' (page header, not the PDF). Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q02","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card lists release channels (Gemini App, Gemini API, Google AI Studio, Gemini for business, Google Antigravity, Google AI Mode) and API access; it gives no release or availability date. Search for the rest of the criterion: Whole card (14 PDF pages). Terms searched: “release”, “launch”, “availab”, “distribut”, “deploy”, “October”. Passages read and not counted: Evaluation Approach, PDF p. 8 (printed p. 7): 'Ethics & Safety Reviews were conducted ahead of the model's release' gives relative timing of a review, not of release. Evaluation, 'Results', PDF p. 5 (printed p. 4): 'results as of October, 2026' dates the results, not the release. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q02-1","mc-ev-cell-google-deepmind-gemini-4-argon-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q03","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the framework with the date 'April 2026' and links the v3.1 document (Frontier Safety Framework v3.1 in the dataset, version date 2026-04-17). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q04","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The PDF's link annotation on p. 11 points to storage.googleapis.com/deepmind-media/DeepMind.com/Blog/strengthening-our-frontier-safety-framework/frontier-safety-framework_3-1.pdf; a link on p. 9 points to deepmind.google/frontier-safety/ (the framework versions page). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q05","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states an outcome for each domain of its table: CBRN uplift TCL not reached for chemical, radiological and nuclear; biological CCL 'cannot be ruled out'; cybersecurity CCL 'cannot be ruled out'; harmful manipulation CCL not reached; stealth and situational awareness TCL 'cannot be ruled out'; ML R&D acceleration and automation CCLs not reached. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q05-1","mc-ev-cell-google-deepmind-gemini-4-argon-q05-2","mc-ev-cell-google-deepmind-gemini-4-argon-q05-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q06","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card ties safety cases (biological and cyber CCLs) and a residual risk assessment with safeguards (stealth and situational awareness TCL) to the thresholds it cannot rule out, but names no safeguard, protection type or safeguard standard; it refers to the Gemini 4 Argon Safety & Security Report. Recorded as partly because Q06 asks for named safeguards. The developer's account, not evidence that the safeguards work. Search for the rest of the criterion: Whole card (Safety & Security; Frontier Safety Assessment and its table; Risks and Mitigations). Terms searched: “safeguard”, “mitigat”, “security level”, “deployment”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q06-1","mc-ev-cell-google-deepmind-gemini-4-argon-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q07","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"CBRN evaluations reported at policy level: the uplift TCL is stated as ruled out for chemical, radiological and nuclear; the biological uplift level 1 CCL is stated as not ruled out. Evaluation details in the row are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q07-1","mc-ev-cell-google-deepmind-gemini-4-argon-q07-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q08","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that the cybersecurity uplift level 1 CCL cannot be ruled out and that safeguards are applied (Q06). Evaluation details are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q09","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The table gives 'CCL not reached' for ML R&D acceleration level 1 and ML R&D automation level 1, on the basis of an internal research engineering benchmark. The loss-of-control (stealth and situational awareness) result is recorded under Q11. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q09-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q10","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card reports human behavioural studies for harmful manipulation and states the level 1 CCL as not reached; it describes the area as one whose evaluations it continues to iterate (Q16). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q10-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q11","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card reports a loss-of-control evaluation of the model's capability for stealth (circumventing basic oversight) and situational awareness, with the stated conclusion that the TCL cannot be ruled out. It reports no propensity evaluation (deception, scheming, sabotage) in the card itself; challenge results are not recorded. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q11-1","mc-ev-cell-google-deepmind-gemini-4-argon-q11-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q12","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states human red teaming by specialist teams in Google's Trust & Safety organization, outside the model development team, covering child safety, content safety policies and issues beyond them, plus automated red teaming; the frontier safety table also refers to expert red teaming in the biological domain. No number of testers or period is given. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q12-1","mc-ev-cell-google-deepmind-gemini-4-argon-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q13","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that its threshold assessments draw on external results but does not name or describe who produced them. Search for the rest of the criterion: Whole card. Terms searched: “external”, “third party”, “third-party”, “partner”, “independent”, “vendor”. Passages read and not counted: Safety & Security, 'Evaluation Approach', PDF p. 8 (printed p. 7): 'developed in partnership with internal safety, security, and responsibility teams': internal teams. Frontier safety table, Loss of Control row, PDF p. 13 (printed p. 12): 'internal and external deployments' refers to deployment settings, not testers. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"partly"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q14","subject":"gemini-4-argon-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Gemini 4 Argon — Model Card was: Whole card. Terms searched: “institute”, “AISI”, “CAISI”, “government”, “UK”, “U.S.”, “external”. Passages read and not counted: Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10): 'external results' names no body (recorded under Q13). The conclusion is bounded by that scope and method.","qualification":"No government body was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q15","subject":"gemini-4-argon-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Gemini 4 Argon — Model Card was: Whole card. Terms searched: “decision”, “decid”, “release”, “launch”, “deploy”, “approv”, “acceptable”. Passages read and not counted: Safety & Security, 'Evaluation Approach', PDF p. 8 (printed p. 7): Evaluations conducted 'to help improve the model and inform decision-making' and Ethics & Safety Reviews 'ahead of the model's release': no decision stated. Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10): Residual risk assessment and safety cases concern risk levels (Q05, Q06), not a release decision. The conclusion is bounded by that scope and method.","qualification":"No statement that a release or deployment decision was made was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-google-deepmind-gemini-4-argon-q16","subject":"gemini-4-argon-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2026-10","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card also states that three thresholds (biological CCL, cybersecurity CCL, stealth and situational awareness TCL) could not be ruled out (Q05) and that the automated safety results are not human evaluation or red teaming. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-google-deepmind-gemini-4-argon-q16-1","mc-ev-cell-google-deepmind-gemini-4-argon-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-meta-llama-2-q01","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “date”, “dated”, “updated”, “version”, “2023”, “2024”. Passages read and not counted: Model Details, 'Model Dates': Training period of the model, not a date of the card. Training Data, 'Data Freshness': Data cut-off, not a date of the card. The conclusion is bounded by that scope and method.","qualification":"The card carries no publication, issue or revision date. Repository metadata (not the card) dates the Llama 2 text at this path to 2023-07-18, with revisions to 2024-03-23 (inventory). A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-meta-llama-2-q02","subject":"llama-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states that Meta released the family and that the models are released openly under a custom commercial licence; it gives no release date or dated event. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-2-q02-1","mc-ev-cell-meta-llama-2-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-meta-llama-2-q06","subject":"llama-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states safety-oriented fine-tuning of the tuned models; no framework determination exists to which it could be tied. The statement is the developer's account. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-2-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-meta-llama-2-q07","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “chemical”, “biolog”, “CBRN”, “weapon”, “uplift”, “hazard”. The conclusion is bounded by that scope and method.","qualification":"No chemical or biological evaluation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-meta-llama-2-q08","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of offensive cyber capability were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “cyber”, “hack”, “exploit”, “malware”, “offensive”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of offensive cyber capability was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q08"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08"}}}},{"id":"mc-cell-meta-llama-2-q09","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “research and development”, “agent”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI research and development was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-meta-llama-2-q10","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “persua”, “manipul”, “influence”, “deceiv”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of persuasion or manipulation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-meta-llama-2-q11","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “decept”, “scheming”, “sabotage”, “misalign”, “honest”, “reward hack”, “oversight”. Passages read and not counted: Evaluation Results, TruthfulQA and ToxiGen tables: Truthfulness and toxicity benchmarks of generated text (content quality and safety), not evaluations of the model's own propensity to deceive or undermine oversight. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-meta-llama-2-q12","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage reporting adversarial testing described as red-teaming (internal, external or both), quoted with its scope (domains, number or kind of testers, or period) where given. Partly if red-teaming is mentioned without any scope. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “red team”, “red-team”, “adversarial”, “jailbreak”. The conclusion is bounded by that scope and method.","qualification":"No passage reporting red-teaming was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q12"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12"}}}},{"id":"mc-cell-meta-llama-2-q13","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “external”, “third-party”, “independent”, “partner”, “evaluator”. Passages read and not counted: Hardware and Software, 'Training Factors': States that evaluation was performed on third-party cloud compute; names no external tester. The conclusion is bounded by that scope and method.","qualification":"No external organisation is named as having tested or evaluated the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-meta-llama-2-q14","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “government”, “institute”, “AISI”, “CAISI”, “AI Safety Institute”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-meta-llama-2-q15","subject":"llama-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Llama 2 model card (MODEL_CARD.md, meta-llama/llama) was: whole card (Markdown file, all sections). Terms searched: “decision”, “decide”, “approv”, “release”, “deploy”. Passages read and not counted: Model Details, first paragraph: States that Meta released the family; no decision or decision-maker is stated. Model Details, 'Status': Refers to future versions to be released; not a decision about this release. The conclusion is bounded by that scope and method.","qualification":"No passage stating a release or deployment decision was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-2-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-meta-llama-2-q16","subject":"llama-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states that its testing was in English and could not cover all scenarios. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-2-q16-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-meta-llama-3-3-q01","subject":"llama-3-3-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of Llama 3.3 model card (MODEL_CARD.md, meta-llama/llama-models) was: whole card (Markdown file, all sections). Terms searched: “date”, “dated”, “updated”, “version”, “2024”, “2025”. Passages read and not counted: Model Information, 'Model Release Date': Release date of the model (Q02), not a date of the card. Training Data, 'Data Freshness': Data cut-off, not a date of the card. The conclusion is bounded by that scope and method.","qualification":"The card carries no publication, issue or revision date of its own; the printed 'Model Release Date' is recorded under Q02. Repository metadata dates the file to 2024-12-06, with revisions on 2024-12-09 and 2025-02-12 (inventory). A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-meta-llama-3-3-q02","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card gives the release date of the 70B Instruct model and states that the models are released openly under the Llama 3.3 Community License. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q02-1","mc-ev-cell-meta-llama-3-3-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-meta-llama-3-3-q06","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states safety fine-tuning and system-level safeguards offered to developers; no framework determination exists to which they could be tied. The statements are the developer's account. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q06-1","mc-ev-cell-meta-llama-3-3-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-meta-llama-3-3-q07","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card reports chemical and biological uplift testing for the Llama 3 family, not Llama 3.3 alone, and states no outcome or conclusion for the domain. Not cross-family checked: the maintainer excluded this cell on chemical and biological evaluations from the cross-family check of 2026-11-03. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"partly"}}}},{"id":"mc-cell-meta-llama-3-3-q08","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card reports a cyber uplift study and an attack-automation study for the Llama 3 family and states no outcome or conclusion. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"partly"}}}},{"id":"mc-cell-meta-llama-3-3-q09","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card reports an attack-automation study of the Llama 3 family as autonomous agents in cyber operations, without an outcome or conclusion. The study is stated for the Llama 3 family, not for Llama 3.3 specifically. No evaluation of self-replication, self-improvement or AI research and development was found. Search for the rest of the criterion: Responsibility & Safety, all subsections; Benchmarks. Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “research and development”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q09-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"partly"}}}},{"id":"mc-cell-meta-llama-3-3-q10","subject":"llama-3-3-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Llama 3.3 model card (MODEL_CARD.md, meta-llama/llama-models) was: whole card (Markdown file, all sections). Terms searched: “persua”, “manipul”, “influence”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of persuasion or manipulation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-meta-llama-3-3-q11","subject":"llama-3-3-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Llama 3.3 model card (MODEL_CARD.md, meta-llama/llama-models) was: whole card (Markdown file, all sections). Terms searched: “decept”, “scheming”, “sabotage”, “misalign”, “honest”, “reward hack”, “oversight”. Passages read and not counted: Responsibility & Safety, 'Refusals and Tone': Refusal behaviour toward benign and borderline prompts (content safety), not misalignment or deception. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-meta-llama-3-3-q12","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states recurring red-teaming and gives the expertise of the red team; it gives no number of testers or period. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q12-1","mc-ev-cell-meta-llama-3-3-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-meta-llama-3-3-q13","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card describes partners who performed red-teaming and subject-matter experts it partnered with, without naming any organisation. Search for the rest of the criterion: Responsibility & Safety, all subsections. Terms searched: “external”, “third-party”, “independent”, “partner”, “vendor”. Passages read and not counted: Responsibility & Safety, 'Community': Membership of the AI Alliance, Partnership on AI and MLCommons is stated; none is named as having tested the model. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"partly"}}}},{"id":"mc-cell-meta-llama-3-3-q14","subject":"llama-3-3-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Llama 3.3 model card (MODEL_CARD.md, meta-llama/llama-models) was: whole card (Markdown file, all sections). Terms searched: “government”, “institute”, “AISI”, “CAISI”, “AI Safety Institute”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-meta-llama-3-3-q15","subject":"llama-3-3-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Llama 3.3 model card (MODEL_CARD.md, meta-llama/llama-models) was: whole card (Markdown file, all sections). Terms searched: “decision”, “decide”, “approv”, “release”, “deploy”. Passages read and not counted: Responsibility & Safety, first paragraph: Describes a 'Responsible release approach'; no decision or decision-maker is stated. Hardware and Software, 'Training Greenhouse Gas Emissions': States that Meta is openly releasing the models; no decision is stated. The conclusion is bounded by that scope and method.","qualification":"No passage stating a release or deployment decision was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-meta-llama-3-3-q16","subject":"llama-3-3-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states that its testing could not cover all scenarios. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-llama-3-3-q16-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q01","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Date printed on the report. Meta's publication page for the report is dated September 24, 2025 (inventory); the report itself does not say whether the October 8 text is a revision. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q02","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report gives the release form (open weights, with open code according to section 1) and no release date. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q02-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-meta-cwm-preparedness-q03","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report names Meta's Frontier AI Framework and cites it with the year 2025, an address and an access date of 2025-09-18, without a version number. As context only: the dataset records the Frontier AI Framework Version 1.1 (dated 2025-02-03) as the edition published before the report's date, and a revised expression under the same label captured from 2025-03-28. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"partly"}}}},{"id":"mc-cell-meta-cwm-preparedness-q04","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The reference list gives the framework's address and an access date. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q05","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states a 'moderate' risk threshold for the framework's catastrophic domains it assessed (cybersecurity, chemical and biological). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q05-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q06","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report describes a prompt-engineering intervention that encourages honest behaviour at inference time, without tying it to the framework determination. Recorded as partly under Q06; the report does not state that the intervention is applied to the release. No safeguard tied to a determination or threshold was found. Search for the rest of the criterion: Abstract; sections 1 to 4 including the Limitations and Summary of results boxes; Appendix A headings. Terms searched: “safeguard”, “mitigat”, “guard”, “protection”, “security standard”, “deployment standard”. Passages read and not counted: Section 1 Introduction, PDF p. 1: 'To anticipate and mitigate risks from this release' introduces the assessment; no safeguard is named. Section 2 Cybersecurity Evaluation, PDF p. 2: Mentions that observed behaviour 'may reflect the effects of safety fine-tuning'; an observation, not a stated safeguard. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-meta-cwm-preparedness-q07","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states that chemical and biological evaluations were run and gives a conclusion for the framework outcomes CB1 and CB2; with the overall 'moderate' threshold statement under Q05. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q08","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states that cybersecurity capability evaluations were run against the framework's Cyber 1 and Cyber 2 scenarios and concludes 'moderate'. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q09","subject":"cwm-preparedness-report","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Code World Model Preparedness Report was: Abstract; sections 1 to 4. Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “research and development”, “machine learning engineering”. Passages read and not counted: Section 4 Propensities, PDF p. 10: Refers to 'greater autonomy' as a motivation for propensity research; no autonomy or AI R&D evaluation is reported. Section 2.2 Agentic Evaluations, PDF pp. 4-5: Agentic tasks are cybersecurity evaluations, recorded under Q08. The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI research and development was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-meta-cwm-preparedness-q10","subject":"cwm-preparedness-report","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Code World Model Preparedness Report was: Abstract; sections 1 to 4. Terms searched: “persua”, “manipul”, “influence”, “social engineering”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of persuasion or manipulation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-meta-cwm-preparedness-q11","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states an honesty (epistemic integrity) evaluation with a conclusion, and states that corrigibility and power-seeking propensities were assessed but their results are not included. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q11-1","mc-ev-cell-meta-cwm-preparedness-q11-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-meta-cwm-preparedness-q12","subject":"cwm-preparedness-report","predicate":"provision_not_found","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage reporting adversarial testing described as red-teaming (internal, external or both), quoted with its scope (domains, number or kind of testers, or period) where given. Partly if red-teaming is mentioned without any scope. The inspected scope of Code World Model Preparedness Report was: Abstract; sections 1 to 4; author list. Terms searched: “red team”, “red-team”, “adversarial”, “jailbreak”. The conclusion is bounded by that scope and method.","qualification":"No passage reporting red-teaming was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q12"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12"}}}},{"id":"mc-cell-meta-cwm-preparedness-q13","subject":"cwm-preparedness-report","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Code World Model Preparedness Report was: Abstract; sections 1 to 4; author list; references. Terms searched: “external”, “third-party”, “independent”, “partner”, “collaboration”. Passages read and not counted: Section 3.1.4 Meta BioKnowledge Proxy, PDF p. 8; section 3.2.2, PDF p. 9: Frontier Design Group and external experts are named as co-developers of two evaluations, not as having tested or evaluated the model. Section 3 and references: SecureBio is cited as the source of benchmarks; not stated to have tested the model. The conclusion is bounded by that scope and method.","qualification":"No external organisation is named as having tested or evaluated the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-meta-cwm-preparedness-q14","subject":"cwm-preparedness-report","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Code World Model Preparedness Report was: Abstract; sections 1 to 4; references. Terms searched: “government”, “institute”, “AISI”, “CAISI”, “AI Safety Institute”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-meta-cwm-preparedness-q15","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states the release and ties it to the assessment, in the first person plural; the report is authored by the 'MSL Preparedness Team' and 'AI Security Team', and no person, office or committee is named as decision-maker. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q15-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"partly"}}}},{"id":"mc-cell-meta-cwm-preparedness-q16","subject":"cwm-preparedness-report","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2025-10-08","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states limits of its cyber evaluations (benchmark coverage, tooling) and that malicious fine-tuning was not included; similar limitation lists follow the chemical and biological and the propensity sections. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-cwm-preparedness-q16-1","mc-ev-cell-meta-cwm-preparedness-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q01","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Date printed on the report. Meta's announcement of Muse Spark 1.3, dated September 2, 2026, links to the report at the same address (inventory); the report does not say whether the October 23 text is a revision. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q02","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report gives the deployment form (Meta API and the Muse product) and no release date. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q02-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q03","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report names the Meta Superintelligence Scaling Framework without a version or date. As context only: the dataset records the Advanced AI Scaling Framework Version 2 (2026-04-08) as the latest Meta edition published before the report's date; a document titled 'Meta Superintelligence Scaling Framework, Version 2.1 (October 2026)' was located at research.meta.ai and is not in the dataset. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"partly"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q04","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report refers to the framework as a document; no address or reference-list entry for it was found. Search for the rest of the criterion: Contents; section 1; references. Terms searched: “Scaling Framework”, “static”, “http”, “Meta Superintelligence Labs”. Passages read and not counted: References: The reference list cites the Muse Spark Safety & Preparedness Report with an address but has no entry for the framework. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"partly"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q05","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states that the 'high risk' threshold is not ruled out before mitigations in the chemical and biological domain, and a residual 'moderate or lower' assessment for chemical and biological and for cybersecurity, and a 'moderate or lower' Loss of Control assessment, for the Muse and Meta API deployments. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q05-1","mc-ev-cell-meta-muse-spark-1-3-q05-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q06","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report ties system-level mitigations to the chemical and biological determination; section 2.1.2 describes them. This is the developer's account, not evidence that the mitigations work. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q07","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states chemical and biological capability evaluations and the framework outcome language before mitigations; the residual assessment is under Q05. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q08","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states cyber capability evaluations against the framework's Cyber 1 and Cyber 2 criteria and a 'moderate or lower' outcome for the unmitigated model. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q09","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states AI R&D evaluations under its Loss of Control domain with a 'moderate or lower' outcome. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q09-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q10","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report places a manipulation (social-engineering) evaluation under the cybersecurity domain; its stated conclusion concerns refusal by the mitigated deployments, not a capability level. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q10-1","mc-ev-cell-meta-muse-spark-1-3-q10-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q11","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states evaluations of strategic underperformance and oversight evasion, with conclusions; further misalignment evaluations are in sections 2.3 and 4. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q11-1","mc-ev-cell-meta-muse-spark-1-3-q11-2","mc-ev-cell-meta-muse-spark-1-3-q11-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q12","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states automated and third-party red-teaming before release, on both deployment variants; sections 3.1.4 and 3.2.2.3 give the jailbreak and prompt-injection scope. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q12-1","mc-ev-cell-meta-muse-spark-1-3-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q13","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report names Handshake AI and Gray Swan as having tested the model; it also reports results on a Scale AI benchmark (section 3.1.4.1). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q13-1","mc-ev-cell-meta-muse-spark-1-3-q13-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q14","subject":"muse-spark-1-3-evaluation-report","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Muse Spark 1.3 Evaluation Report was: Sections 1 to 3; section headings of 4; references. Terms searched: “government”, “institute”, “AISI”, “CAISI”, “AI Safety Institute”, “Security Institute”. Passages read and not counted: Section 2.1.1.1, PDF p. 16: '(US AISI and UK AISI, 2024)' is cited for a prior evaluation practice; neither body is stated to have tested or received access to Muse Spark 1.3. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q15","subject":"muse-spark-1-3-evaluation-report","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Muse Spark 1.3 Evaluation Report was: Section 1 (including 1.2 Governance); section 2.3.1; searches across the report. Terms searched: “decision”, “decide”, “approv”, “release”, “launch”, “Chief AI Officer”, “Director of Alignment”. Passages read and not counted: Section 1.2 Governance, PDF pp. 11-12: States that the framework includes 'oversight and decisionmaking processes'; no decision about this model or its maker is stated. Section 2.3.1, PDF p. 34: 'Reliable pre-deployment evaluations help ensure deployment decisions are properly informed' is a general statement, not a decision. The conclusion is bounded by that scope and method.","qualification":"No passage stating a release or deployment decision was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-meta-muse-spark-1-3-q16","subject":"muse-spark-1-3-evaluation-report","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2026-10-23","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The report states limits of individual assessments; several sections carry their own 'Limitations' paragraphs. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-meta-muse-spark-1-3-q16-1","mc-ev-cell-meta-muse-spark-1-3-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-2-q01","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “date”, “dated”, “updated”, “version”, “2023”, “2024”, “2025”. The conclusion is bounded by that scope and method.","qualification":"The card carries no date. Repository metadata dates the README to 2023-12-13, with revisions to 2025-12-08 (inventory). A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-microsoft-phi-2-q02","subject":"phi-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card gives the release form (open source, MIT licence) and no release date. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q02-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-microsoft-phi-2-q06","subject":"phi-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states safety-motivated filtering of training data and that the model was not fine-tuned with reinforcement learning from human feedback; no framework determination exists to which it could be tied. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-microsoft-phi-2-q07","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “chemical”, “biolog”, “CBRN”, “weapon”, “hazard”. The conclusion is bounded by that scope and method.","qualification":"No chemical or biological evaluation was found. The search is recorded below. Not cross-family checked: the maintainer excluded this cell on chemical and biological evaluations from the cross-family check of 2026-11-03. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-microsoft-phi-2-q08","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of offensive cyber capability were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “cyber”, “hack”, “exploit”, “malware”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of offensive cyber capability was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q08"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08"}}}},{"id":"mc-cell-microsoft-phi-2-q09","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “agent”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI research and development was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-microsoft-phi-2-q10","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “persua”, “manipul”, “influence”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of persuasion or manipulation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-microsoft-phi-2-q11","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “decept”, “scheming”, “sabotage”, “misalign”, “honest”, “oversight”. Passages read and not counted: Limitations of Phi-2, 'Toxicity' and 'Potential Societal Biases': Limitations of model outputs, not evaluations of misalignment or deception. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-microsoft-phi-2-q12","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage reporting adversarial testing described as red-teaming (internal, external or both), quoted with its scope (domains, number or kind of testers, or period) where given. Partly if red-teaming is mentioned without any scope. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “red team”, “red-team”, “adversarial”, “jailbreak”. The conclusion is bounded by that scope and method.","qualification":"No passage reporting red-teaming was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q12"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12"}}}},{"id":"mc-cell-microsoft-phi-2-q13","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “external”, “third-party”, “independent”, “partner”, “evaluator”. The conclusion is bounded by that scope and method.","qualification":"No external organisation is named as having tested or evaluated the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-microsoft-phi-2-q14","subject":"phi-2-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Phi-2 model card (README.md, microsoft/phi-2) was: whole card (Markdown file, all sections). Terms searched: “government”, “institute”, “AISI”, “CAISI”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-microsoft-phi-2-q15","subject":"phi-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states a choice to release the model, with its reason, in the company's voice ('we'). No person, office or committee is named. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q15-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-2-q16","subject":"phi-2-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states a gap in its testing for production use. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-2-q16-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-4-q01","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “date”, “dated”, “updated”, “version”. Passages read and not counted: Model Summary, 'Release date': Release date of the model (Q02). Model Summary, 'Dates': Training period, not a date of the card. The conclusion is bounded by that scope and method.","qualification":"The card carries no publication, issue or revision date of its own; the printed 'Release date' is recorded under Q02. Repository metadata dates the README to 2024-12-11, with revisions to 2025-11-24 (inventory). A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-microsoft-phi-4-q02","subject":"phi-4-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card gives a release date and the release form (open model, MIT licence). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q02-1","mc-ev-cell-microsoft-phi-4-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-4-q06","subject":"phi-4-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states safety post-training; no framework determination exists to which it could be tied. The statement is the developer's account. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-microsoft-phi-4-q07","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “chemical”, “biolog”, “CBRN”, “weapon”, “hazard”. The conclusion is bounded by that scope and method.","qualification":"No chemical or biological evaluation was found. The search is recorded below. Not cross-family checked: the maintainer excluded this cell on chemical and biological evaluations from the cross-family check of 2026-11-03. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-microsoft-phi-4-q08","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of offensive cyber capability were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “cyber”, “hack”, “exploit”, “malware”. Passages read and not counted: Responsible AI Considerations, 'Misuse': States that malware production may be possible; a stated limitation, not an evaluation. The conclusion is bounded by that scope and method.","qualification":"No evaluation of offensive cyber capability was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q08"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08"}}}},{"id":"mc-cell-microsoft-phi-4-q09","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “agent”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI research and development was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-microsoft-phi-4-q10","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “persua”, “manipul”, “influence”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of persuasion or manipulation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-microsoft-phi-4-q11","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “decept”, “scheming”, “sabotage”, “misalign”, “honest”, “oversight”. Passages read and not counted: Data Overview, Training Datasets, item 4: Mentions training data reflecting preferences for honesty; not an evaluation. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-microsoft-phi-4-q12","subject":"phi-4-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states red-teaming by Microsoft's AI Red Team, with its scope (average and adversarial user scenarios, single- and multi-turn interactions, jailbreak techniques). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q12-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-4-q13","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “external”, “third-party”, “independent”, “partner”, “evaluator”. Passages read and not counted: Safety, Safety Evaluation and Red-Teaming: The 'independent AI Red Team (AIRT) at Microsoft' is a Microsoft team, not an external organisation. The conclusion is bounded by that scope and method.","qualification":"No external organisation is named as having tested or evaluated the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-microsoft-phi-4-q14","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “government”, “institute”, “AISI”, “CAISI”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-microsoft-phi-4-q15","subject":"phi-4-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Phi-4 model card (README.md, microsoft/phi-4) was: whole card (Markdown file, all sections). Terms searched: “decision”, “decide”, “approv”, “release”, “deploy”. Passages read and not counted: Safety, Safety Evaluation and Red-Teaming: 'Prior to release, `phi-4` followed a multi-faceted evaluation approach' describes evaluation before release; no decision or decision-maker is stated. The conclusion is bounded by that scope and method.","qualification":"No passage stating a release or deployment decision was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-microsoft-phi-4-q16","subject":"phi-4-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states that the model was not evaluated for all downstream purposes. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-q16-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q01","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “date”, “dated”, “updated”, “version”. Passages read and not counted: Training, Model, 'Release date': Release date of the model (Q02). Training, Model, 'Dates': Training period, not a date of the card. The conclusion is bounded by that scope and method.","qualification":"The card carries no publication, issue or revision date of its own; the printed 'Release date' is recorded under Q02. Repository metadata dates the README to 2025-06-22, with revisions to 2025-12-10 (inventory). A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q02","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card gives a release month and the release form (open model, MIT licence; also offered in Azure AI Foundry). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q02-1","mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q03","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card names the developer’s framework (Responsible Scaling Policy, Preparedness Framework, Frontier Safety Framework, Risk Management Framework, Frontier AI Framework, Advanced AI Scaling Framework, Frontier Governance Framework, or a successor). Stated if it names the framework and identifies the edition (version number, date or a link to a dated edition); partly if it names the framework without an edition. Not applicable if the card predates every framework edition of its developer. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “framework”, “Frontier AI Framework”, “Frontier Governance”, “Scaling Framework”, “policy”, “threshold”, “governance”, “Frontier Governance Framework”, “tracked”. Passages read and not counted: Safety Evaluation and Red-Teaming: 'Azure AI Foundry’s Risk and Safety Evaluation framework' is an evaluation tool, not Microsoft's frontier framework. Safety Evaluation and Red-Teaming: 'Microsoft's responsible AI principles' are principles, not the frontier framework. The conclusion is bounded by that scope and method.","qualification":"The card postdates Microsoft's Frontier Governance Framework of February 2025 and does not name it. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q03"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q04","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A hyperlink, address or bibliographic reference to the framework document in the card. Partly if it refers to the framework as a document without an address or reference. Not applicable if Q03 is not applicable. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “framework”, “Frontier Governance”, “http”, “aka.ms”. The conclusion is bounded by that scope and method.","qualification":"No link or reference to a framework document was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q04"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q05","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A sentence stating the outcome for the model in the framework’s own categories (for example, a safety level applied, a capability threshold reached or not reached, a risk category assigned), quoted. Partly if the card says a determination was made without stating its outcome, or states an outcome for some tracked domains only. Not applicable if Q03 is not applicable. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “threshold”, “level”, “determination”, “tracked”, “high-risk capabilit”. Passages read and not counted: Safety Evaluation and Red-Teaming: 'Risk and Safety Evaluation framework' names an evaluation tool, not a frontier safety framework. The conclusion is bounded by that scope and method.","qualification":"The card postdates Microsoft's Frontier Governance Framework (February 2025) but names no framework (Q03); no outcome in a framework's categories was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q05"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q06","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states safety post-training, described for the Phi-4 family; no framework determination is stated to which it is tied. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q07","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “chemical”, “biolog”, “CBRN”, “weapon”, “hazard”. The conclusion is bounded by that scope and method.","qualification":"No chemical or biological evaluation was found. The search is recorded below. Not cross-family checked: the maintainer excluded this cell on chemical and biological evaluations from the cross-family check of 2026-11-03. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q08","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of offensive cyber capability were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “cyber”, “hack”, “exploit”, “malware”. Passages read and not counted: Responsible AI Considerations, 'Misuse': States that malware production may be possible; a stated limitation, not an evaluation. The conclusion is bounded by that scope and method.","qualification":"No evaluation of offensive cyber capability was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q08"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q09","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “agent”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI research and development was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q10","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “persua”, “manipul”, “influence”. Passages read and not counted: Responsible AI Considerations, 'Election Information Reliability': A stated product limitation, not an evaluation of persuasion. The conclusion is bounded by that scope and method.","qualification":"No evaluation of persuasion or manipulation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q11","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “decept”, “scheming”, “sabotage”, “misalign”, “honest”, “oversight”. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q12","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"Red-teaming appears in the section heading; the section's text describes an automated risk and safety evaluation and does not describe red-teaming or its scope. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q12-1","mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"partly"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q13","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “external”, “third-party”, “independent”, “partner”, “evaluator”. The conclusion is bounded by that scope and method.","qualification":"No external organisation is named as having tested or evaluated the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q14","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “government”, “institute”, “AISI”, “CAISI”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q15","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Phi-4-mini-flash-reasoning model card (README.md, microsoft/Phi-4-mini-flash-reasoning) was: whole card (Markdown file, all sections). Terms searched: “decision”, “decide”, “approv”, “release”, “deploy”. Passages read and not counted: Release Notes: 'This release ... addresses user feedback and market demand for a compact reasoning model' gives a reason for the release; no decision or decision-maker is stated. The conclusion is bounded by that scope and method.","qualification":"No passage stating a release or deployment decision was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q16","subject":"phi-4-mini-flash-reasoning-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states the limited scope of its testing. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q16-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q01","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A date printed on the card or its own page (publication, issue or revision date), quoted with its kind. Revision dates are listed separately. Partly if the card carries only a month or year, or only a revision date. The release date of the model is Q02. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “date”, “dated”, “updated”, “version”, “2026”. Passages read and not counted: Model summary table, 'Release date', PDF p. 1: Release date of the model (Q02). The conclusion is bounded by that scope and method.","qualification":"The card carries no publication, issue or revision date of its own; the printed release date is recorded under Q02. The PDF's internal metadata (not printed) records creation on 2026-10-27. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q01"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q02","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card gives a release date and a restricted release form (private preview within MDASH, subject to approval). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q02-1","mc-ev-cell-microsoft-mai-cyber-1-1-flash-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q03","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card names the developer’s framework (Responsible Scaling Policy, Preparedness Framework, Frontier Safety Framework, Risk Management Framework, Frontier AI Framework, Advanced AI Scaling Framework, Frontier Governance Framework, or a successor). Stated if it names the framework and identifies the edition (version number, date or a link to a dated edition); partly if it names the framework without an edition. Not applicable if the card predates every framework edition of its developer. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “framework”, “Frontier AI Framework”, “Frontier Governance”, “Scaling Framework”, “policy”, “threshold”, “governance”, “Frontier Governance Framework”, “tracked”, “high-risk capabilit”. Passages read and not counted: Section 1.2 Alignment approach, PDF p. 2: 'Microsoft Responsible AI Principles and Microsoft policies' are named; not the frontier framework. The conclusion is bounded by that scope and method.","qualification":"The card postdates Microsoft's Frontier Governance Framework editions of February 2025 and February 2026 and does not name the framework. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q03"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q04","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A hyperlink, address or bibliographic reference to the framework document in the card. Partly if it refers to the framework as a document without an address or reference. Not applicable if Q03 is not applicable. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “framework”, “Frontier Governance”, “http”, “aka.ms”. The conclusion is bounded by that scope and method.","qualification":"No link or reference to a framework document was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q04"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q05","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A sentence stating the outcome for the model in the framework’s own categories (for example, a safety level applied, a capability threshold reached or not reached, a risk category assigned), quoted. Partly if the card says a determination was made without stating its outcome, or states an outcome for some tracked domains only. Not applicable if Q03 is not applicable. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “threshold”, “level”, “determination”, “tracked”, “critical”, “high-risk”. Passages read and not counted: Section 2.8 Responsible AI considerations, PDF p. 4: 'Because of the increased model-level risk' states a risk judgement without framework categories; recorded under Q06. The conclusion is bounded by that scope and method.","qualification":"The card names no frontier framework (Q03); no outcome in a framework's categories was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q05"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q06","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card ties access restrictions and system safeguards to its own risk findings, not to a framework determination. The statement is the developer's account, not evidence that the safeguards work. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q07","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in this domain were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion for the domain. Radiological and nuclear evaluations are noted where reported together. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “chemical”, “biolog”, “CBRN”, “weapon”, “radiolog”, “nuclear”. The conclusion is bounded by that scope and method.","qualification":"No chemical or biological evaluation was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q07"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q08","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card reports offensive-cyber capability evaluations and states that an internal red-team assessment found 'critical risks of the model alone'; it uses no framework threshold language. Benchmark details are not recorded here. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q08-1","mc-ev-cell-microsoft-mai-cyber-1-1-flash-q08-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q09","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “autonom”, “self-replic”, “self-improv”, “AI R&D”, “research and development”. Passages read and not counted: Section 1.1 Key model capabilities, PDF p. 2: Agentic security workflows are described as product capability. The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI research and development was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q10","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card names a manipulation evaluation among its safety measurements and reports no result for it. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q10-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10","status":"partly"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q11","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “decept”, “scheming”, “sabotage”, “misalign”, “alignment”, “oversight”, “monitor”. Passages read and not counted: Section 1.2 Alignment approach, PDF p. 2: Training on safety data and monitoring during cyber evaluations are stated; no evaluation of misalignment or deception is reported. The conclusion is bounded by that scope and method.","qualification":"No evaluation of misalignment, deception, scheming or sabotage was found. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q12","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states red-teaming by the MAI internal red team (38 scenarios), by Microsoft's AI Red Team within MDASH and by an external firm (Q13). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q12-1","mc-ev-cell-microsoft-mai-cyber-1-1-flash-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q13","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card names Casaba Security as an external evaluator. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q14","subject":"mai-cyber-1-1-flash-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of MAI-Cyber-1.1-Flash model card was: whole card (8 PDF pages). Terms searched: “government”, “institute”, “AISI”, “CAISI”, “agency”. The conclusion is bounded by that scope and method.","qualification":"No government body is named as having tested or received access to the model. The search is recorded below. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q15","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states a deployment decision (access limited to MDASH) without naming who made it. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q15-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"partly"}}}},{"id":"mc-cell-microsoft-mai-cyber-1-1-flash-q16","subject":"mai-cyber-1-1-flash-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"unknown","note":"The card prints no date of its own, so its cells carry no date."},"qualification":"The card states the scope limit of its benchmark results. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-microsoft-mai-cyber-1-1-flash-q16-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-openai-gpt-4v-q01","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card prints September 25, 2023 beneath its title and author line. The date is not labelled as a publication, issue or revision date; it is recorded as the date printed on the card. No revision date was located. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-openai-gpt-4v-q02","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card dates early access to March 2023 and states that GPT-4V is being made broadly available. It gives no date for broad availability other than its own date. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q02-1","mc-ev-cell-openai-gpt-4v-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-openai-gpt-4v-q06","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states model-level and system-level mitigations. They are not tied to a framework determination or threshold; the card predates OpenAI’s framework editions. Stated safeguards are the developer’s account, not evidence that they work. Search for the rest of the criterion: §2.4 Mitigations; §3 Conclusion and Next Steps. Terms searched: “mitigation”, “safeguard”, “refusal”, “moderation”, “threshold”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q06-1","mc-ev-cell-openai-gpt-4v-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-openai-gpt-4v-q07","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that GPT-4V inherits the GPT-4 System Card’s assessment of CBRN risk areas without restating its outcome, and that image input does not meaningfully alter capability in these categories. Red teamers’ scientific-proficiency testing touched on dangerous chemical synthesis; the card states the outputs were inaccurate and error-prone. No evaluation run on GPT-4V with an outcome category or threshold language is reported. Details of the red-team example are not recorded (hazard rule). Search for the rest of the criterion: §2.2 Evaluations; §2.3 External Red Teaming incl. footnote 2; §2.3.1 Scientific proficiency. Terms searched: “CBRN”, “chemical”, “biolog”, “weapon”, “nuclear”, “radiolog”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q07-1","mc-ev-cell-openai-gpt-4v-q07-2","mc-ev-cell-openai-gpt-4v-q07-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"partly"}}}},{"id":"mc-cell-openai-gpt-4v-q08","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card lists CAPTCHA-breaking accuracy evaluations under “Cybersecurity” and states that such a capability can have cybersecurity implications. No outcome or conclusion for these evaluations is stated in the card’s text layer (figure contents are not captured). No other offensive-cyber evaluation is reported. Search for the rest of the criterion: §2.2 Evaluations; §2.3; Appendix A. Terms searched: “cyber”, “CAPTCHA”, “exploit”, “hacking”, “vulnerab”. Passages read and not counted: Footnote 2, PDF p. 6: Refers to “system interaction” assessed in the GPT-4 System Card; no cyber outcome stated for GPT-4V. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q08-1","mc-ev-cell-openai-gpt-4v-q08-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"partly"}}}},{"id":"mc-cell-openai-gpt-4v-q09","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that GPT-4V inherits the GPT-4 System Card’s assessment of emergent risky properties such as self-replication, without restating its outcome; no autonomy or AI R&D evaluation of GPT-4V is reported. Search for the rest of the criterion: Whole card. Terms searched: “self-replication”, “autonom”, “agent”, “power-seeking”, “R&D”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q09-1","mc-ev-cell-openai-gpt-4v-q09-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"partly"}}}},{"id":"mc-cell-openai-gpt-4v-q10","subject":"gpt-4v-system-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of GPT-4V(ision) System Card was: Whole card, especially §2.3.4 Disinformation risks. Terms searched: “persua”, “manipulat”, “influence”, “convinc”, “disinformation”. Passages read and not counted: §2.3.4 Disinformation risks, PDF pp. 9–11: Concerns generating text tailored to images and the model’s ability to detect disinformation; it cites prior studies on belief in image-paired statements and reports no evaluation of the model’s persuasive or manipulative capability. §2.3.3 Stereotyping and ungrounded inferences, PDF p. 9: Concerns biased inferences about people, not persuasion. The conclusion is bounded by that scope and method.","qualification":"No passage reporting an evaluation of persuasive or manipulative capability was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-openai-gpt-4v-q11","subject":"gpt-4v-system-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of GPT-4V(ision) System Card was: Whole card. Terms searched: “decept”, “deceiv”, “scheming”, “sabotag”, “misalign”, “oversight”, “reward hack”, “power-seeking”, “alignment”. Passages read and not counted: Footnote 2, PDF p. 6: Refers to “emergent risky properties such as self-replication” assessed in the GPT-4 System Card; recorded under Q09. §2.2 and §2.4, refusal evaluations: Content-safety refusal tests, which the criterion excludes. The conclusion is bounded by that scope and method.","qualification":"No passage reporting an evaluation of misalignment, deception, scheming or sabotage of oversight was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-openai-gpt-4v-q12","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"External red teaming focused on the vision functionality; the card names six key risk areas (scientific proficiency, medical advice, stereotyping and ungrounded inferences, disinformation risks, hateful content, visual vulnerabilities). The acknowledgements name 25 individual red teamers. No period is given. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q12-1","mc-ev-cell-openai-gpt-4v-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-openai-gpt-4v-q13","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names Be My Eyes as an early-access organisation whose pilot of a GPT-4V-based product it describes as testing. External red teamers are named in the acknowledgements as individuals, without organisations. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-openai-gpt-4v-q14","subject":"gpt-4v-system-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of GPT-4V(ision) System Card was: Whole card. Terms searched: “government”, “institute”, “AISI”, “AI Safety Institute”, “NIST”, “Department”. Passages read and not counted: §2.2 Person identification evaluations, PDF p. 4: “public servants and politicians” and a dataset of images of members of Congress are evaluation data, not government testers. The conclusion is bounded by that scope and method.","qualification":"No passage naming a government body that tested or received access to the model was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-openai-gpt-4v-q15","subject":"gpt-4v-system-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of GPT-4V(ision) System Card was: §1 Introduction; §2 Deployment Preparation; §3 Conclusion and Next Steps. The cross-family second pass of 2026-11-03 read the whole card again (printed pp. 1 and 14 cited). Terms searched: “decision”, “decide”, “decided”, “approv”, “chose”, “deemed”, “acceptable”, “determine”, “therefore”, “deploy”, “release”, “launch”. Passages read and not counted: §1 Introduction, PDF p. 1: “This system card outlines how OpenAI prepared the vision capabilities of GPT-4 for deployment.” Deployment preparation, not a release decision. §1 Introduction, PDF p. 1: “is the latest capability we are making broadly available”. States availability; no decision, assessment or maker is described. The conclusion is bounded by that scope and method.","qualification":"The card states that OpenAI is making GPT-4V broadly available and describes how it prepared the model for deployment. It does not describe a decision to release, say that the release follows from an assessment, or name who decided. Recorded as not found under the question’s criterion as written (maintainer decision 13a, settled when the cross-family second pass of 2026-11-03 held the checker’s reading). Maintainer decision 13a of 2026-11-03. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-openai-gpt-4v-q16","subject":"gpt-4v-system-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2023-09-25","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that red teaming is unlikely to have found all visual vulnerabilities and that parity on its demographic evaluations does not rule out differing downstream harms. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-4v-q16-1","mc-ev-cell-openai-gpt-4v-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-openai-o1-q01","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Printed cover date December 5, 2024; footnote 2 states the card was released on December 5th. Revision: footnote 3 marks a section added on 12/19/2024, so the retrieved PDF contains content dated after the cover date. The dataset’s record of the web version carries the label “Updated: December 5, 2024”. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q01-1","mc-ev-cell-openai-o1-q01-2","mc-ev-cell-openai-o1-q01-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-openai-o1-q02","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card dates the launch of an improved o1 to December 17th (the card itself is dated December 5, 2024) and refers to deployment of o1 on the API. It gives no separate date for the initial availability of o1. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q02-1","mc-ev-cell-openai-o1-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-openai-o1-q03","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card names the Preparedness Framework and cites it as reference [28], titled “(beta)”, 2023, with the address cdn.openai.com/openai-preparedness-framework-beta.pdf; the words “Preparedness Framework” in §5 (PDF p. 15) also carry a hyperlink to that address. Edition: the Preparedness Framework (Beta). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q03-1","mc-ev-cell-openai-o1-q03-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-openai-o1-q04","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Bibliographic reference to the framework, with the printed address https://cdn.openai.com/openai-preparedness-framework-beta.pdf (accessed 2024-09-11; the text layer separates the address prefix from the rest of the address). The phrase “The Preparedness Framework is a living document” in §5 (PDF p. 15) carries a hyperlink to the same PDF. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-openai-o1-q05","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states the Safety Advisory Group’s classification for all four tracked categories, pre- and post-mitigation. The classification passage is also quoted in the dataset’s AI and biosecurity records. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q05-1","mc-ev-cell-openai-o1-q05-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-openai-o1-q06","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card ties strengthened mitigations and enhanced monitoring to the Medium designations for CBRN and persuasion. This is the developer’s account, not evidence that the safeguards work. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q06-1","mc-ev-cell-openai-o1-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"stated"}}}},{"id":"mc-cell-openai-o1-q07","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Chemical and biological: Medium. Radiological and nuclear are reported in a separate section, with a Low designation for the post-mitigation model. Evaluation methods, tasks and results are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q07-1","mc-ev-cell-openai-o1-q07-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-openai-o1-q08","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Cybersecurity: Low. The card reports capture-the-flag evaluations; results are not recorded here. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-openai-o1-q09","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Model autonomy: Low. The card also reports METR’s evaluation of an earlier checkpoint on agentic tasks (§4.4.4), with METR’s caveats. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q09-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-openai-o1-q10","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Persuasion: Medium, below the High threshold. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q10-1","mc-ev-cell-openai-o1-q10-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10","status":"stated"}}}},{"id":"mc-cell-openai-o1-q11","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card reports Apollo Research’s scheming evaluations with Apollo’s stated conclusion, and OpenAI’s own chain-of-thought deception monitor results. The conclusions are as the card reports them. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q11-1","mc-ev-cell-openai-o1-q11-2","mc-ev-cell-openai-o1-q11-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-openai-o1-q12","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"External red teaming by multiple organisations and individuals, October to early December 2024, across the listed categories; it includes a pairwise comparison by members of the Red Teaming Network (§4.4.1) and a jailbreak arena with Gray Swan (§4.4.2). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q12-1","mc-ev-cell-openai-o1-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"stated"}}}},{"id":"mc-cell-openai-o1-q13","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"Named external evaluators include Apollo Research, METR and Gray Swan; the card also names Signature Science (§5.5.10) among organisations engaged for expert probing. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q13-1","mc-ev-cell-openai-o1-q13-2","mc-ev-cell-openai-o1-q13-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-openai-o1-q14","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card names the US AISI and the “UK Safety Institute (UK AISI)” as having conducted pre-deployment evaluations of a version of o1; their results are not included. The passage is also quoted in the dataset’s AI and biosecurity records. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q14-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14","status":"stated"}}}},{"id":"mc-cell-openai-o1-q15","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card names the Safety Advisory Group as the maker of the risk classification (Q05) and states the framework’s deployment condition. It refers to the deployment of the models but does not state who decided to release them. Search for the rest of the criterion: §5; §5.1; §6 Conclusion. Terms searched: “decision”, “decide”, “approv”, “Safety Advisory Group”, “leadership”, “Board”, “deploy”, “launch”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q15-1","mc-ev-cell-openai-o1-q15-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"partly"}}}},{"id":"mc-cell-openai-o1-q16","subject":"o1-card-2024","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2024-12-19","basis":"revision","note":"Dated as the card read is dated: the date of its revision, as the source records it."},"qualification":"The card states that its Preparedness evaluations are a lower bound and that its radiological and nuclear assessment is limited by what could be tested. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-o1-q16-1","mc-ev-cell-openai-o1-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q01","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Date printed on the cover, not labelled by kind. No revision date was located in the card. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q02","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card gives the form of deployment (in Codex, not in general-purpose chat). No release or availability date is given apart from the card’s own date. Search for the rest of the criterion: §1; §2.1; §3 Product-Specific Risk Mitigations. Terms searched: “release”, “launch”, “available”, “deploy”, “API”, “today”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q02-1","mc-ev-cell-openai-gpt-5-2-codex-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q03","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the Preparedness Framework without a version number in its text. The word “Preparedness” in the second passage carries a hyperlink to cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf, which identifies the edition: preparedness-v2. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q03-1","mc-ev-cell-openai-gpt-5-2-codex-q03-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q04","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Hyperlinks (PDF link annotations): on PDF p. 12 to the Preparedness Framework Version 2 PDF; on PDF p. 8 to openai.com/index/updating-our-preparedness-framework/ (OpenAI’s announcement of the updated framework). The reference list does not cite the framework. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q04-1","mc-ev-cell-openai-gpt-5-2-codex-q04-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q05","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Biological and Chemical: treated as High. Cybersecurity: below High. AI self-improvement: below High. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q05-1","mc-ev-cell-openai-gpt-5-2-codex-q05-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q06","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card ties deployment with the GPT-5 family’s safeguards to the High biology treatment; §5.1.1 repeats that the corresponding safeguards continue to apply. This is the developer’s account, not evidence that the safeguards work. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q07","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card reports four biological and chemical evaluations; outcome category: treated as High. Evaluation contents and results are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q08","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Cybersecurity: below High. The card reports three internal benchmarks and an external evaluation by Irregular; results are not recorded here. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q09","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"AI self-improvement: below High, with four evaluations reported. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q09-1","mc-ev-cell-openai-gpt-5-2-codex-q09-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q10","subject":"gpt-5-2-codex-system-card-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Addendum to GPT-5.2 System Card: GPT-5.2-Codex was: Whole card (22 pages). Terms searched: “persua”, “manipulat”, “influence”, “convinc”. Passages read and not counted: §5.1.2.3 Cyber Range, Table 7, PDF p. 16 (printed p. 15): “LDAP manipulation” names a cyber-range technique, not manipulation of people. The conclusion is bounded by that scope and method.","qualification":"No passage reporting an evaluation of persuasive or manipulative capability was found. The card’s Preparedness section addresses the Biological and Chemical, Cybersecurity and AI Self-Improvement domains. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q11","subject":"gpt-5-2-codex-system-card-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports evaluations of the model’s own propensity or capability for deception, scheming, reward hacking, sabotage or undermining oversight, with a stated conclusion. Partly if evaluations are reported without a conclusion. Content-safety refusal tests do not meet it. The inspected scope of Addendum to GPT-5.2 System Card: GPT-5.2-Codex was: Whole card. Terms searched: “decept”, “scheming”, “sabotag”, “sandbag”, “reward hack”, “misalign”, “oversight”, “alignment”, “monitor”. Passages read and not counted: §4.2 Avoid Data-Destructive Actions, PDF p. 8: Evaluates avoidance of destructive actions and preservation of user changes; it does not address deception, scheming or undermining oversight. §5.1.2 Cybersecurity, PDF p. 11 (printed p. 10): “monitoring and enforcement pipeline” concerns misuse by users, not the model’s own conduct. The conclusion is bounded by that scope and method.","qualification":"No passage reporting an evaluation of deception, scheming, reward hacking, sabotage or undermining oversight was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q11"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q12","subject":"gpt-5-2-codex-system-card-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage reporting adversarial testing described as red-teaming (internal, external or both), quoted with its scope (domains, number or kind of testers, or period) where given. Partly if red-teaming is mentioned without any scope. The inspected scope of Addendum to GPT-5.2 System Card: GPT-5.2-Codex was: Whole card. Terms searched: “red team”, “red-team”, “red teaming”, “adversarial”, “external”. Passages read and not counted: §2.2 Jailbreaks, PDF p. 5 (printed p. 4): An automated jailbreak benchmark (StrongReject adaptation), not described as red-teaming. §5.1.2.4 External Evaluations by Irregular, PDF pp. 16–17: An external capability evaluation, not described as red-teaming; recorded under Q13. The conclusion is bounded by that scope and method.","qualification":"No passage describing adversarial testing as red-teaming was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q12"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q13","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Irregular is named as an external evaluator of cyber capability. SecureBio and Gryphon Scientific are named as sources of evaluation datasets, not as testers of this model. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q13-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q14","subject":"gpt-5-2-codex-system-card-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Addendum to GPT-5.2 System Card: GPT-5.2-Codex was: Whole card. Terms searched: “government”, “institute”, “CAISI”, “AISI”, “AI Safety Institute”, “AI Security Institute”, “national”. The conclusion is bounded by that scope and method.","qualification":"No passage naming a government body that tested or received access to the model was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q15","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the Safety Advisory Group and OpenAI leadership as having determined the cyber capability level, and states that the model is being deployed. It does not state who decided to release the model. Search for the rest of the criterion: §1; §5. Terms searched: “decision”, “decide”, “approv”, “Safety Advisory Group”, “leadership”, “deploy”, “launch”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q15-1","mc-ev-cell-openai-gpt-5-2-codex-q15-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"partly"}}}},{"id":"mc-cell-openai-gpt-5-2-codex-q16","subject":"gpt-5-2-codex-system-card-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2025-12-18","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card lists limitations of each cyber benchmark and states that results on them are not sufficient to establish High cyber capability. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-5-2-codex-q16-1","mc-ev-cell-openai-gpt-5-2-codex-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q01","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Date printed on the cover (2026-10-29), not labelled by kind; the title also names October 29, 2026. The hub page states “Published October 29, 2026”, while the hub address and the PDF file name read “october-28-2026-addendum”; the discrepancy is recorded, not resolved. No revision date was located. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q02","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card introduces the updated model “today” (the card is dated 2026-10-29) and dates the release of the earlier GPT-6 Luna in ChatGPT to October 7. §8.2 refers to the updated model’s “public launch”. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q02-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q03","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the Preparedness Framework without a version in its text. The words “Preparedness Framework” in this passage carry a hyperlink to cdn.openai.com/pdf/18a02b5d-6b67-4cec-ab64-68cdfbddebcd/preparedness-framework-v2.pdf, which identifies the edition: preparedness-v2. The card does not name the Frontier Governance Framework. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q03-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q04","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Hyperlink (PDF link annotation) on “Preparedness Framework,” to the Preparedness Framework Version 2 PDF. No bibliographic reference to the framework. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q05","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Cybersecurity and Biological and Chemical: treated as High. AI Self-Improvement: below High. The card states these match the assessments for GPT-6 Luna (October 7). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q05-1","mc-ev-cell-openai-gpt-6-luna-october-29-q05-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q06","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Safeguards are tied to the High determinations and described by reference to the October 7 card; §8.2 is a public summary of an internal Safeguards Report. This is the developer’s account, not evidence that the safeguards work. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q06-1","mc-ev-cell-openai-gpt-6-luna-october-29-q06-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q07","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Outcome category: High. The card reports updated results against indicative High and Critical thresholds; results are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q08","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Outcome category: High. Results of the two reported evaluations are not recorded here. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q09","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states the outcome (below High) but reports no AI self-improvement evaluation run on the October 29 model; it refers to the GPT-6 Astra card for the evaluations. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q09-1","mc-ev-cell-openai-gpt-6-luna-october-29-q09-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09","status":"partly"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q10","subject":"gpt-6-luna-october-29-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Addendum to GPT-6 Luna System Card: October 29, 2026 Update was: Whole card (16 pages). Terms searched: “persua”, “manipulat”, “influence”, “convinc”, “agreeable”. Passages read and not counted: §1 Introduction, PDF p. 4: “reduces overly agreeable language” describes a change in the model’s style, not an evaluation of persuasion or manipulation. The conclusion is bounded by that scope and method.","qualification":"No passage reporting an evaluation of persuasive or manipulative capability was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q11","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card’s Alignment section reports four evaluations (respecting auto-review, respecting warnings, blocker deception, broken search tool) as figures. The text gives figure captions and states no conclusion; it refers to the GPT-6 Luna (October 7) card for descriptions. Search for the rest of the criterion: §7 Alignment, PDF pp. 10–13. Terms searched: “decept”, “scheming”, “sabotag”, “misalign”, “oversight”, “circumvent”, “monitor”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q11-1","mc-ev-cell-openai-gpt-6-luna-october-29-q11-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"partly"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q12","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card describes an automated red-teaming agent used in adversarial training, and automated 'GPT-Red' evaluations of direct and indirect prompt injection. It does not describe adversarial testing of this model as red-teaming, and reports no human red-teaming. Search for the rest of the criterion: Whole card. Terms searched: “red team”, “red-team”, “adversarial”, “attacker”, “jailbreak”. Passages read and not counted: §4.1 Jailbreaks, PDF pp. 6–7: Reports defender success rates against an automated attacker; not described as red-teaming. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q12-1","mc-ev-cell-openai-gpt-6-luna-october-29-q12-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12","status":"partly"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q13","subject":"gpt-6-luna-october-29-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Addendum to GPT-6 Luna System Card: October 29, 2026 Update was: Whole card. Terms searched: “external”, “third-party”, “third party”, “partner”, “evaluated by”, “Irregular”, “Apollo”, “METR”, “SecureBio”. Passages read and not counted: §4.2 Prompt Injection, PDF p. 7: “malicious instructions in third-party content” describes an attack type, not a tester. The conclusion is bounded by that scope and method.","qualification":"No passage naming an external tester or evaluator was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q14","subject":"gpt-6-luna-october-29-addendum","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Addendum to GPT-6 Luna System Card: October 29, 2026 Update was: Whole card. Terms searched: “government”, “institute”, “CAISI”, “AISI”, “AI Safety Institute”, “AI Security Institute”. The conclusion is bounded by that scope and method.","qualification":"No passage naming a government body that tested or received access to the model was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q15","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states a determination by OpenAI leadership, informed by a Safety Advisory Group recommendation, that safeguards are sufficient for the model’s public launch. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q15-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15","status":"stated"}}}},{"id":"mc-cell-openai-gpt-6-luna-october-29-q16","subject":"gpt-6-luna-october-29-addendum","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2026-10-29","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states that the jailbreak evaluation run was incomplete and that comparison values may not match those published at launch. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-openai-gpt-6-luna-october-29-q16-1","mc-ev-cell-openai-gpt-6-luna-october-29-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q01","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card prints a full date under its title without a label (publication or issue not specified). No revision date is printed. The file name (2025-11-17-grok-4-1-model-card.pdf) carries the same date. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q02","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card gives the form of availability (consumer web and mobile apps) in the present tense, with no release date. Search for the rest of the criterion: Whole card (6 pages). Terms searched: “release”, “launch”, “availab”, “deploy”, “API”. Passages read and not counted: §1 Introduction, p. 1: 'pre-deployment safety testing' gives no date. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q02-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-1-q03","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names xAI's Risk Management Framework; its reference gives the year 2025 only, with no version, full date or address. The RMF is not among the dataset's framework editions (open question). Context only, not the card's statement: the RMF served at data.x.ai prints 'Last updated: August 20, 2025'. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q03-1","mc-ev-cell-xai-grok-4-1-q03-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-1-q04","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"A bibliographic reference to the framework (author, title, year), cited in §2 as '[xAI, 2025]'; it gives no address. The PDF's only link annotation points to the Grok 4 model card. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q05","subject":"grok-4-1-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A sentence stating the outcome for the model in the framework’s own categories (for example, a safety level applied, a capability threshold reached or not reached, a risk category assigned), quoted. Partly if the card says a determination was made without stating its outcome, or states an outcome for some tracked domains only. Not applicable if Q03 is not applicable. The inspected scope of Grok 4.1 Model Card was: Whole card. Terms searched: “threshold”, “RMF”, “Risk Management Framework”, “level”, “acceptable”, “determin”. Passages read and not counted: §1 Introduction and §2 Evaluations, p. 1: RMF named as the basis for grouping evaluations; no outcome stated. §2.3.2 Results, p. 5: 'we do not believe it poses risks due to advanced persuasion capabilities' is a conclusion (Q10) not expressed in the framework's categories. The conclusion is bounded by that scope and method.","qualification":"No determination in the framework's own categories (a threshold reached or not reached) was found; the card groups its evaluations by the RMF's three risk categories and reports results and conclusions outside threshold language. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q05"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05"}}}},{"id":"mc-cell-xai-grok-4-1-q06","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card ties input filters to its chemical and biological evaluation results, not to a framework determination or threshold (none is stated, Q05). It also describes refusal training and input filters in §2.1.1. The developer's account, not evidence that the safeguards work. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-1-q07","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Biology and chemistry dual-use evaluations reported, with a stated comparative conclusion and comparisons with human baselines; no threshold language. The section's scope sentence names radiological and nuclear weapons, but the evaluations listed cover biology, chemistry and cyber only. Scores are not recorded (hazard rule). Not cross-family checked: the maintainer excluded this cell on chemical and biological evaluations from the cross-family check of 2026-11-03. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q08","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Dual-use cyber knowledge and an agentic capture-the-flag evaluation are reported, with the stated comparative conclusion quoted (the card's wording); no threshold language. Scores are not recorded. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q09","subject":"grok-4-1-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Grok 4.1 Model Card was: Whole card. Terms searched: “autonom”, “agentic”, “self-”, “replicat”, “AI R&D”, “research and development”, “loss of control”. Passages read and not counted: §2.1.2 Evaluations, p. 2: Agentic refusal and prompt-injection evaluations measure misuse refusal and robustness, not autonomy. §2.3.1 Evaluations, p. 4: The agentic capture-the-flag evaluation is a cyber evaluation (Q08). The conclusion is bounded by that scope and method.","qualification":"No evaluation of autonomy, self-replication, self-improvement or AI R&D was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-xai-grok-4-1-q10","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"A persuasion evaluation (MakeMeSay) is reported with the stated conclusion quoted. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q10-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q11","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card reports a deception (lying under pressure) evaluation and a sycophancy evaluation as 'concerning propensities', with the conclusion quoted; its table gives dishonesty rates for Grok 4 and both Grok 4.1 configurations. No scheming, sabotage or oversight evaluation was found. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q11-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-1-q12","subject":"grok-4-1-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage reporting adversarial testing described as red-teaming (internal, external or both), quoted with its scope (domains, number or kind of testers, or period) where given. Partly if red-teaming is mentioned without any scope. The inspected scope of Grok 4.1 Model Card was: Whole card. Terms searched: “red team”, “red-team”, “red teaming”, “adversarial”, “jailbreak”, “tester”. Passages read and not counted: §2.1.2 Evaluations, 'Adversarial robustness', p. 2: Adversarial testing on internal jailbreak templates and a prompt-injection suite; not described as red-teaming. §2.1.1 Safety Training Approach, p. 1: Adversarial attacks used to train input filters; training, not testing described as red-teaming. The conclusion is bounded by that scope and method.","qualification":"No testing described as red-teaming was found. The card reports adversarial robustness evaluations (jailbreak templates, prompt injection) without calling them red-teaming. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q12"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12"}}}},{"id":"mc-cell-xai-grok-4-1-q13","subject":"grok-4-1-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. At least one external organisation (other than a government body, which is Q14) named as having tested or evaluated the model, quoted. Partly if external testers are described but not named. The inspected scope of Grok 4.1 Model Card was: Whole card, including references. Terms searched: “external”, “third party”, “third-parties”, “partner”, “independent”, “evaluator”, “RAND”. Passages read and not counted: §2.3.1 Evaluations and References, pp. 4-6: Published benchmarks and human baselines by other authors (including a RAND report) are cited as sources, not as testers of Grok 4.1. §3.1 Data and Training, p. 5: 'data produced by third-parties' concerns training data. The conclusion is bounded by that scope and method.","qualification":"No external organisation named as having tested or evaluated Grok 4.1 was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q13"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13"}}}},{"id":"mc-cell-xai-grok-4-1-q14","subject":"grok-4-1-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Grok 4.1 Model Card was: Whole card. Terms searched: “institute”, “AISI”, “CAISI”, “government”, “UK”, “U.S.”. The conclusion is bounded by that scope and method.","qualification":"No government body was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-xai-grok-4-1-q15","subject":"grok-4-1-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Grok 4.1 Model Card was: Whole card. Terms searched: “decision”, “decid”, “release”, “launch”, “deploy”, “approv”. Passages read and not counted: §1 Introduction, p. 1: 'we engage in pre-deployment safety testing' and 'We also deploy these models with safeguards' describe testing and deployment conditions; no decision or decision-maker is stated. The conclusion is bounded by that scope and method.","qualification":"No statement that a release or deployment decision was made was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-xai-grok-4-1-q16","subject":"grok-4-1-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2025-11-17","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states an error in earlier cards' refusal results, that the new multilingual results are not directly comparable, and that the human baselines used may underestimate expert performance. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-1-q16-1","mc-ev-cell-xai-grok-4-1-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q01","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Card date"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The cover prints the card's date (kind not labelled) and a revision date, both 2026-09-21. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q01-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q01","facet":{"id":"MC.Q01","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q02","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Release date"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card lists release channels (SpaceXAI API, Grok Build, Cursor, Office add-ins, model gateways) and a planned consumer release 'at a later date'; it gives no release date. The consumer release is announced, not stated as made. Search for the rest of the criterion: Whole card (30 PDF pages). Terms searched: “release”, “launch”, “availab”, “deploy”, “September”. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q02-1","mc-ev-cell-xai-grok-4-7-q02-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q02","facet":{"id":"MC.Q02","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-7-q03","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework cited"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names the framework and its reference gives the edition date June 30, 2026 (dataset faif-2026). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q03-1","mc-ev-cell-xai-grok-4-7-q03-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q03","facet":{"id":"MC.Q03","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q04","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Link to framework"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The reference carries an address (media.x.ai/v1/website/xai-frontier-artificial-intelligence-framework-30-june-2026-99c40684.pdf), also as a link annotation. That address returned HTTP 404 on 2026-11-02; the dataset's faif-2026 record names it as the file formerly served for the June 2026 edition. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q04-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q04","facet":{"id":"MC.Q04","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q05","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Framework determination"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card states an outcome in the framework's terms for dual-use biological and chemical knowledge only; no FAIF threshold outcome is stated for cyber or other domains. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q05-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q05","facet":{"id":"MC.Q05","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-7-q06","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Safeguards tied to determination"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card describes a layered safeguard stack (safety training, system prompts, runtime input and topical filters on some deployment surfaces, cyber-specific safeguards) without tying it to the FAIF result; it attributes the biological result to training environments and data selection rather than to safeguards. The developer's account, not evidence that the safeguards work. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q06-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q06","facet":{"id":"MC.Q06","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-7-q07","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Chemical and biological"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Biological and chemical capability evaluations are reported as FAIF safety-threshold evaluations, run without production safeguards, with the threshold language quoted. Radiological and nuclear items appear only in a refusal evaluation (§9.3). Scores are not recorded (hazard rule). What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q07-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q07","facet":{"id":"MC.Q07","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q08","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Cyber"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"Offensive cyber capability evaluations are reported, run without production safeguards, with a comparative conclusion; no FAIF threshold language is given for cyber. The card states that third-party evaluators corroborated the internal cyber results (Q13). Scores are not recorded. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q08-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q08","facet":{"id":"MC.Q08","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q09","subject":"grok-4-7-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Autonomy and AI R&D"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations in any of these domains were run and states their outcome category or threshold language (policy level only). Partly if evaluations are reported without any outcome or conclusion. Agentic task benchmarks reported only as product capability do not meet it. The inspected scope of Model Card: Grok 4.7 was: Whole card. Terms searched: “autonom”, “self-repl”, “replicat”, “self-improv”, “AI R&D”, “AI research”, “loss of control”, “oversight”. Passages read and not counted: §1 Introduction and §1.1 Overview, PDF p. 4 (printed p. 3): Autonomy described as capability on coding, engineering and office work. §2 Coding capabilities, PDF pp. 6-11: Long-horizon agentic coding benchmarks, one including AI research challenges, reported as product capability. The conclusion is bounded by that scope and method.","qualification":"No dangerous-capability evaluation of autonomy, self-replication, self-improvement or AI R&D was found; the card's autonomy statements concern product capability. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q09"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q09","facet":{"id":"MC.Q09"}}}},{"id":"mc-cell-xai-grok-4-7-q10","subject":"grok-4-7-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Persuasion"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. The card reports that evaluations of persuasive or manipulative capability were run and states their outcome category or conclusion. Partly if evaluations are reported without an outcome. Not found includes a card that states the domain was removed from evaluation; the note says so. The inspected scope of Model Card: Grok 4.7 was: Whole card. Terms searched: “persua”, “manipul”, “influence”. The conclusion is bounded by that scope and method.","qualification":"No persuasion or manipulation evaluation was found; the terms do not occur in the card. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q10"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q10","facet":{"id":"MC.Q10"}}}},{"id":"mc-cell-xai-grok-4-7-q11","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Alignment and deception"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card reports an honesty-under-pressure evaluation and a sycophancy evaluation with rates in tables, without a concluding sentence. No scheming, sabotage or oversight evaluation was found. Stated, partly: the passage has what the question’s criterion asks for, in a weaker or narrower form than the criterion’s own words. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q11-1"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q11","facet":{"id":"MC.Q11","status":"partly"}}}},{"id":"mc-cell-xai-grok-4-7-q12","subject":"grok-4-7-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Red-teaming"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage reporting adversarial testing described as red-teaming (internal, external or both), quoted with its scope (domains, number or kind of testers, or period) where given. Partly if red-teaming is mentioned without any scope. The inspected scope of Model Card: Grok 4.7 was: Whole card. Terms searched: “red team”, “red-team”, “red-teaming”, “adversarial”, “jailbreak”. Passages read and not counted: §6.3 and §6.4, PDF pp. 18-19 (printed pp. 17-18): 'red-teaming tasks' and 'a red-team cyber evaluation' measure the model's offensive cyber capability (Q08), not adversarial testing of the model. §7.1 BioSecBench, PDF p. 20 (printed p. 19): 'red-team refusal' names a refusal measure within a capability suite; no red-teaming exercise is described. §8 Jailbreaks and robustness, PDF p. 23 (printed p. 22): Adversarial testing of production safeguards with jailbreak attacks; not described as red-teaming. The conclusion is bounded by that scope and method.","qualification":"No adversarial testing of the model described as red-teaming was found. The card's uses of 'red-team' describe offensive cyber tasks the model performs; its adversarial testing is described as jailbreak testing. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q12"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q12","facet":{"id":"MC.Q12"}}}},{"id":"mc-cell-xai-grok-4-7-q13","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"External testers"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card names Datacurve (and, in a similar note on PDF p. 9, Harbor) as having run evaluations, describes CathedralBench as a third-party evaluation, and thanks nine 'evaluation partners' (Abundant AI, Datacurve, Mecado, Atopile, Harbor, Proximal Labs, Cathedral, LatchBio, Vals AI). The third-party evaluators given unrestricted cyber access are not named. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q13-1","mc-ev-cell-xai-grok-4-7-q13-2","mc-ev-cell-xai-grok-4-7-q13-3"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q13","facet":{"id":"MC.Q13","status":"stated"}}}},{"id":"mc-cell-xai-grok-4-7-q14","subject":"grok-4-7-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Government institute"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A named government body (for example, the US AI Safety Institute or Center for AI Standards and Innovation, or the UK AI Safety or Security Institute) stated to have tested, evaluated or received access to the model, quoted. Partly if government testing is mentioned without naming the body. The inspected scope of Model Card: Grok 4.7 was: Whole card, including references and acknowledgements. Terms searched: “institute”, “AISI”, “CAISI”, “government”, “UK”, “U.S.”. Passages read and not counted: References, item 5, PDF p. 29 (printed p. 28): 'Stanford & Laude Institute' are named as benchmark authors, not a government body. The conclusion is bounded by that scope and method.","qualification":"No government body was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q14"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q14","facet":{"id":"MC.Q14"}}}},{"id":"mc-cell-xai-grok-4-7-q15","subject":"grok-4-7-model-card","predicate":"provision_not_found","object":{"kind":"literal","value":"Release decision"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"inferenceRationale":"Premises: the model-card question asks for the following. A passage stating that a decision to release, deploy or launch was made (decided, chose, deemed acceptable for deployment, determined that safeguards suffice for launch), quoted. Stated if the decision is described and its maker is identifiable: a named person, office or committee, or the developer in its own voice (“we”), which counts as the company as such. Partly if a release is stated as following from an assessment without a decision being described (“we therefore release”), if a decision is described without any maker, or if a maker is named only for a capability or safeguards determination and not for the release. The inspected scope of Model Card: Grok 4.7 was: Whole card. Terms searched: “decision”, “decid”, “release”, “launch”, “deploy”, “approv”. Passages read and not counted: §1.1 Overview, PDF p. 4 (printed p. 3): 'final deployed checkpoint' identifies the evaluated checkpoint; 'not intended for autonomous high-stakes decision-making' is a usage limitation. Neither states a release decision. §1.1 Overview, PDF p. 5: Availability channels and a planned consumer release, stated without a decision. The conclusion is bounded by that scope and method.","qualification":"No statement that a release or deployment decision was made was found. A recorded absence within the stated scope of the card read: not evidence that the developer did not do the thing asked about, or did not disclose it elsewhere.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q15"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q15","facet":{"id":"MC.Q15"}}}},{"id":"mc-cell-xai-grok-4-7-q16","subject":"grok-4-7-model-card","predicate":"states_provision","object":{"kind":"literal","value":"Stated limits"},"topic":"governance","time":{"kind":"event","date":"2026-09-21","basis":"publication","note":"Dated as the card read is dated: its printed date, as the source records it."},"qualification":"The card also states that a live benchmark suite should be frozen to the evaluated snapshot and that some benchmark versions are not comparable with earlier ones. What the card states, quoted from the expression read. A card is the developer’s own account: not evidence that an evaluation was run as described, that a safeguard works, or that a finding is independent.","evidenceIds":["mc-ev-cell-xai-grok-4-7-q16-1","mc-ev-cell-xai-grok-4-7-q16-2"],"attributes":{"provision":{"kind":"disclosure","key":"model-card-q16","facet":{"id":"MC.Q16","status":"stated"}}}}],"sources":[{"id":"source-o1-card","documentId":"o1-card-2024","publisher":"OpenAI","publisherId":"openai"},{"id":"source-eval-i-claude2-card","documentId":"claude2-model-card","publisher":"Anthropic","publisherId":"anthropic"},{"id":"mc-src-claude-3-5-addendum","documentId":"claude-3-5-haiku-sonnet-addendum-2024","publisher":"Anthropic","publisherId":"anthropic"},{"id":"mc-src-claude-opus-4-5-card","documentId":"claude-opus-4-5-system-card","publisher":"Anthropic","publisherId":"anthropic"},{"id":"mc-src-claude-haiku-5-5-card","documentId":"claude-haiku-5-5-system-card","publisher":"Anthropic","publisherId":"anthropic"},{"id":"mc-src-gemini-3-flash-card","documentId":"gemini-3-flash-model-card","publisher":"Google DeepMind","publisherId":"google-deepmind"},{"id":"mc-src-gemini-4-argon-card","documentId":"gemini-4-argon-model-card","publisher":"Google DeepMind","publisherId":"google-deepmind"},{"id":"mc-src-llama-2-card","documentId":"llama-2-model-card","publisher":"Meta","publisherId":"meta"},{"id":"mc-src-llama-3-3-card","documentId":"llama-3-3-model-card","publisher":"Meta","publisherId":"meta"},{"id":"mc-src-cwm-preparedness-report","documentId":"cwm-preparedness-report","publisher":"Meta","publisherId":"meta"},{"id":"mc-src-muse-spark-1-3-report","documentId":"muse-spark-1-3-evaluation-report","publisher":"Meta","publisherId":"meta"},{"id":"mc-src-phi-2-card","documentId":"phi-2-model-card","publisher":"Microsoft","publisherId":"microsoft"},{"id":"mc-src-phi-4-card","documentId":"phi-4-model-card","publisher":"Microsoft","publisherId":"microsoft"},{"id":"mc-src-phi-4-mini-flash-reasoning-card","documentId":"phi-4-mini-flash-reasoning-model-card","publisher":"Microsoft","publisherId":"microsoft"},{"id":"mc-src-mai-cyber-1-1-flash-card","documentId":"mai-cyber-1-1-flash-model-card","publisher":"Microsoft","publisherId":"microsoft"},{"id":"mc-src-gpt-4v-card","documentId":"gpt-4v-system-card","publisher":"OpenAI","publisherId":"openai"},{"id":"mc-src-gpt-5-2-codex-addendum","documentId":"gpt-5-2-codex-system-card-addendum","publisher":"OpenAI","publisherId":"openai"},{"id":"mc-src-gpt-6-luna-october-29-addendum","documentId":"gpt-6-luna-october-29-addendum","publisher":"OpenAI","publisherId":"openai"},{"id":"mc-src-grok-4-1-card","documentId":"grok-4-1-model-card","publisher":"xAI","publisherId":"xai"},{"id":"mc-src-grok-4-7-card","documentId":"grok-4-7-model-card","publisher":"xAI","publisherId":"xai"}],"evidence":[{"id":"mc-ev-cell-anthropic-claude-2-1-q01-1","claimId":"mc-cell-anthropic-claude-2-1-q01","sourceId":"mc-src-claude-2-1-card","locator":"Section 1 'Introduction', PDF p. 1","passage":"Update (November 21, 2023): We have added an appendix section covering Claude 2.1.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q02-1","claimId":"mc-cell-anthropic-claude-2-1-q02","sourceId":"mc-src-claude-2-1-card","locator":"Section 2 'Claude 2 Model Card', 'Model details', PDF p. 1","passage":"Claude 2 was developed by Anthropic and released in July 2023.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q02-2","claimId":"mc-cell-anthropic-claude-2-1-q02","sourceId":"mc-src-claude-2-1-card","locator":"Claude 2.1 Model Card Appendix, Section 1 'Introduction', PDF p. 15","passage":"Claude 2.1, a new release which builds additional product features and functionality on top of Claude 2.0","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q03","claimId":"mc-cell-anthropic-claude-2-1-q03","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q04","claimId":"mc-cell-anthropic-claude-2-1-q04","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q05","claimId":"mc-cell-anthropic-claude-2-1-q05","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q06-1","claimId":"mc-cell-anthropic-claude-2-1-q06","sourceId":"mc-src-claude-2-1-card","locator":"Section 2 'Evaluations and Red Teaming', PDF p. 2","passage":"we do not believe any deployed versions of Claude pose national security or significant safety related risks in the areas that we have identified … partly to mitigations that we have put in place","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q07","claimId":"mc-cell-anthropic-claude-2-1-q07","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q08","claimId":"mc-cell-anthropic-claude-2-1-q08","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q09-1","claimId":"mc-cell-anthropic-claude-2-1-q09","sourceId":"mc-src-claude-2-1-card","locator":"Section 2 'Evaluations and Red Teaming', PDF p. 2","passage":"Neither ARC nor we believe that our current Claude models possess the dangerous capabilities (‘autonomous replication’ abilities) that ARC is aiming to detect","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q10","claimId":"mc-cell-anthropic-claude-2-1-q10","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q11","claimId":"mc-cell-anthropic-claude-2-1-q11","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q12-1","claimId":"mc-cell-anthropic-claude-2-1-q12","sourceId":"mc-src-claude-2-1-card","locator":"Section 2 'Evaluations and Red Teaming', PDF p. 2","passage":"Before deployment, we also worked with external red teamers, including crowdworker platforms, to test Claude 2 on a range of Trust and Safety related topics","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q12-2","claimId":"mc-cell-anthropic-claude-2-1-q12","sourceId":"mc-src-claude-2-1-card","locator":"Section 2 'Evaluations and Red Teaming', PDF p. 2","passage":"We evaluated and red-teamed Claude 2 and previous Claude models for several national security and safety related risks.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q13-1","claimId":"mc-cell-anthropic-claude-2-1-q13","sourceId":"mc-src-claude-2-1-card","locator":"Section 2 'Evaluations and Red Teaming', PDF p. 2","passage":"We have been working with the Alignment Research Center (ARC) since fall of 2022 to support their safety audits of our AI models.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q14","claimId":"mc-cell-anthropic-claude-2-1-q14","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q15","claimId":"mc-cell-anthropic-claude-2-1-q15","sourceId":"mc-src-claude-2-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-2-1-q16-1","claimId":"mc-cell-anthropic-claude-2-1-q16","sourceId":"mc-src-claude-2-1-card","locator":"Section 3.1 'Human Feedback Evaluations and Red-Teaming', PDF p. 3","passage":"This approach has its limitations–for instance, we know the scenarios created by crowdworkers are not fully representative of the scenarios Claude will encounter out in real-world usage","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-2-1-q16-2","claimId":"mc-cell-anthropic-claude-2-1-q16","sourceId":"mc-src-claude-2-1-card","locator":"Appendix Section 3 'Honesty Evaluations', PDF p. 16","passage":"Our internal evaluations for this metric are still in early stages of development","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q01","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q01","sourceId":"mc-src-claude-3-5-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q02-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q02","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 2.3 'Vision Capabilities', PDF p. 5","passage":"Claude 3.5 Haiku will initially launch as a text-only model","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q02-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q02","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2 'Frontier Risk Evaluations', PDF p. 9","passage":"we conducted comprehensive safety evaluations on the upgraded Claude 3.5 Sonnet prior to release","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q03-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q03","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 1 'Introduction', PDF p. 1","passage":"Both models underwent extensive safety evaluations, including comprehensive testing for potential risks in biological, cybersecurity, and autonomous behavior domains, in accordance with our Responsible Scaling Policy (RSP) [4].","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q03-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q03","sourceId":"mc-src-claude-3-5-addendum","locator":"References [4], PDF p. 13 (the address breaks across two lines)","passage":"[4] Anthropic, “Responsible Scaling Policy.” October, 2024. https://assets.anthropic.com/m/ 24a47b00f10301cd/original/Anthropic-Responsible-Scaling-Policy-2024-10-15.pdf.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q04-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q04","sourceId":"mc-src-claude-3-5-addendum","locator":"References [4], PDF p. 13 (the address breaks across two lines)","passage":"[4] Anthropic, “Responsible Scaling Policy.” October, 2024. https://assets.anthropic.com/m/ 24a47b00f10301cd/original/Anthropic-Responsible-Scaling-Policy-2024-10-15.pdf.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q05-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q05","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2.1 'Results', PDF p. 10","passage":"Based on our analysis, we judge that the upgraded Claude 3.5 Sonnet does not require ASL-3 safeguards at this time.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q05-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q05","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.3 'Ongoing Safety Commitment', PDF p. 10","passage":"While these models did not trigger the full evaluation protocol described in our Responsible Scaling Policy","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q06-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q06","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 2.1 'Computer Use', PDF p. 3","passage":"As with all our models, we implement safeguards and continue to monitor for potential misuse.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q06-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q06","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.1.3 'Computer Use Red-Teaming', PDF p. 9","passage":"we are establishing several monitoring protocols and mitigations to ensure a safe and measured release of this capability","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q07-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q07","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3 'Safety', PDF p. 7","passage":"We also evaluate frontier risks in accordance with our Responsible Scaling Policy (RSP) [4] in the the areas of Chemical, Biological, Radiological, and Nuclear (CBRN) risks, Cybersecurity, and Autonomous Capabilities.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q07-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q07","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2.1 'Results', PDF pp. 9–10 (the text layer splits 'ASL-3' across the page break; the ellipsis marks the page number)","passage":"The model did not demonstrate capabilities requiring ASL … 3 safeguards and security in any domain.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q08-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q08","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3 'Safety', PDF p. 7","passage":"We also evaluate frontier risks in accordance with our Responsible Scaling Policy (RSP) [4] in the the areas of Chemical, Biological, Radiological, and Nuclear (CBRN) risks, Cybersecurity, and Autonomous Capabilities.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q08-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q08","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2.1 'Results', PDF pp. 9–10 (the text layer splits 'ASL-3' across the page break; the ellipsis marks the page number)","passage":"The model did not demonstrate capabilities requiring ASL … 3 safeguards and security in any domain.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q09-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q09","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.1.4 'Evaluating Computer Use for the Responsible Scaling Policy', PDF p. 9","passage":"we deem our software engineering evaluations as currently sufficient for ruling out ASL-3 autonomy capabilities","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q09-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q09","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2.1 'Results', PDF p. 10","passage":"Based on our analysis, we judge that the upgraded Claude 3.5 Sonnet does not require ASL-3 safeguards at this time.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q10","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q10","sourceId":"mc-src-claude-3-5-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q11","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q11","sourceId":"mc-src-claude-3-5-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q12-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q12","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.1.1 'Model Red-Teaming', PDF p. 7","passage":"We conducted comprehensive Trust & Safety (T&S) evaluations across fourteen policy areas in six languages","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q12-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q12","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.1.3 'Computer Use Red-Teaming', PDF p. 8","passage":"We conducted specific Trust & Safety red-teaming for computer use to identify potential abuse vectors.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q12-3","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q12","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 2.1 'Computer Use', PDF p. 2","passage":"We conducted rigorous first party and independent third party multimodal red-teaming exercises","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q13-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q13","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 1 'Introduction', PDF p. 1; repeated in Section 3.2, PDF p. 9","passage":"We also collaborated with METR [8] to conduct an independent assessment.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q14-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q14","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 1 'Introduction', PDF p. 1; repeated in Section 3.2, PDF p. 9","passage":"joint pre-deployment testing of the new Claude 3.5 Sonnet model was conducted by the US AI Safety Institute (US AISI) [6] and the UK AI Safety Institute (UK AISI) [7].","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q15-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q15","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2.1 'Results', PDF p. 10","passage":"Based on our analysis, we judge that the upgraded Claude 3.5 Sonnet does not require ASL-3 safeguards at this time.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q15-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q15","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.2 'Frontier Risk Evaluations', PDF p. 9","passage":"we conducted comprehensive safety evaluations on the upgraded Claude 3.5 Sonnet prior to release","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q16-1","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q16","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 2.2 'Tool Use & Agentic Tasks', PDF p. 3","passage":"We believe dedicated scaffolding and prompting can further improve the results.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-3-5-haiku-sonnet-new-q16-2","claimId":"mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q16","sourceId":"mc-src-claude-3-5-addendum","locator":"Section 3.1.4 'Evaluating Computer Use for the Responsible Scaling Policy', PDF p. 9","passage":"We detail some of our preliminary reasoning below","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q01-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q01","sourceId":"mc-src-claude-opus-4-5-card","locator":"Cover, PDF p. 1","passage":"System Card: Claude Opus 4.5 November 2025","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q01-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q01","sourceId":"mc-src-claude-opus-4-5-card","locator":"Changelog, PDF p. 2","passage":"November 24, 2025 … November 25, 2025 … December 5, 2025","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q02-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q02","sourceId":"mc-src-claude-opus-4-5-card","locator":"Abstract, PDF p. 3","passage":"Informed by the testing described here, we have deployed Claude Opus 4.5 under the AI Safety Level 3 Standard.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q02-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q02","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 3.2 'Ambiguous context evaluations', PDF p. 41","passage":"In preparation for this model launch, we were able to reduce this behavior by modifying the system prompt that is applied for conversations on Claude.ai.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q03-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q03","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.3 'AI Safety Level determination process', PDF p. 12 (hyperlinked words)","passage":"For more information on the requirements for these ASL Standards, please see the Responsible Scaling Policy","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q04-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q04","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.3 'AI Safety Level determination process', PDF p. 12 (hyperlinked words)","passage":"For more information on the requirements for these ASL Standards, please see the Responsible Scaling Policy","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q05-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q05","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.4 'Conclusions', PDF p. 12","passage":"Our determination is that Claude Opus 4.5 does not cross either the AI R&D-4 or CBRN-4 capability threshold.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q06-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q06","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.1 'Overview', PDF p. 11","passage":"For Claude Opus 4.5, we implemented ASL-3 (AI Safety Level 3) protections based on the model’s demonstrated capabilities.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q06-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q06","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.3 'AI Safety Level determination process', PDF p. 12","passage":"Based on these assessments, we have decided to release Claude Opus 4.5 under the ASL-3 Standard.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q07-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q07","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.4.2 'On chemical, biological, radiological, and nuclear (CBRN) risks', PDF p. 14","passage":"We determine that Claude Opus 4.5 does not cross the CBRN-4 threshold.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q07-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q07","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.2 'CBRN evaluations', 'Claude Opus 4.5 Results', PDF p. 119","passage":"As a result, we determined ASL-3 safeguards were appropriate.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q07-3","claimId":"mc-cell-anthropic-claude-opus-4-5-q07","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.2.1 'On chemical risks', PDF p. 120","passage":"we do not currently run specific evaluations on chemical risks internally","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q08-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q08","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.4 'Cyber evaluations', 'Results', PDF p. 145","passage":"We believe that Claude Opus 4.5 does not demonstrate catastrophically risky capabilities in the cyber domain.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q08-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q08","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.4 'Cyber evaluations', PDF p. 143","passage":"The Responsible Scaling Policy does not stipulate a capability threshold for cyber capabilities at any ASL level","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q09-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q09","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.4.1 'On autonomy risks', PDF p. 13","passage":"We judge that Claude Opus 4.5 could not fully automate an entry-level, remote-only research role at Anthropic.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q09-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q09","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.3 'AI Safety Level determination process', PDF p. 12","passage":"Claude Opus 4.5 has roughly reached the pre-defined thresholds we set for straightforward ASL-4 rule-out based on benchmark tasks.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q10-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q10","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 6.1 'Introduction and summary of findings', PDF p. 64","passage":"attempts to hide dangerous capabilities, and attempts to manipulate users toward certain views","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q11-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q11","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 6.1.2 'Overall assessment of high-stakes sabotage risk', PDF p. 66","passage":"Claude Opus 4.5 is very unlikely to pose a significant risk of catastrophic impacts through the intentional sabotage of human engineering, R&D, or decision-making.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q11-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q11","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 6.1.1 'Key findings on safety and alignment', PDF p. 66","passage":"We saw no signs of sandbagging or strategic attempts to tamper with our safety testing in a way that would affect our RSP determinations","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q12-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q12","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.2.4.10 'ASL-4 red teaming with the CAISI', PDF p. 132","passage":"We worked with the US Center for AI Standards and Innovation to red-team Claude Opus 4.5 over a one-week window, which included at least 2–3 days of red-teaming.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q12-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q12","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.2 'CBRN evaluations', 'Claude Opus 4.5 Results', PDF p. 119","passage":"Finally, red-teaming results by various third parties demonstrated some increase in risk, but not enough to require additional protections.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q13-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q13","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 5.2.1 'Gray Swan Agent Red Teaming benchmark for tool use', PDF p. 58","passage":"Gray Swan measured the success rate of prompt injection attacks after a single attempt (k=1)","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q14-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q14","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.5 'Third party assessments', PDF p. 151","passage":"pre-deployment testing of Claude Opus 4.5 was conducted by the US Center for AI Standards and Innovation (CAISI) and the UK AI Security Institute (UK AISI).","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q14-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q14","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 7.2.2 'On radiological and nuclear risks', PDF p. 120","passage":"Anthropic has maintained a formal partnership with the U.S. Department of Energy’s National Nuclear Security Administration (NNSA) to evaluate our AI models for potential nuclear and radiological risks.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q15-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q15","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.3 'AI Safety Level determination process', PDF p. 12","passage":"Consistent with our RSP, the Responsible Scaling Officer and CEO made the ultimate determination on the required ASL Standards.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q15-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q15","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.3 'AI Safety Level determination process', PDF p. 12","passage":"Based on these assessments, we have decided to release Claude Opus 4.5 under the ASL-3 Standard.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q16-1","claimId":"mc-cell-anthropic-claude-opus-4-5-q16","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.4 'Conclusions', PDF p. 12","passage":"However, confidently ruling out these thresholds is becoming increasingly difficult.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-opus-4-5-q16-2","claimId":"mc-cell-anthropic-claude-opus-4-5-q16","sourceId":"mc-src-claude-opus-4-5-card","locator":"Section 1.2.4 'Conclusions', PDF p. 12","passage":"parts of the AI R&D-4 and CBRN-4 thresholds have fundamental epistemic uncertainty or require more sophisticated forms of measurement","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q01-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q01","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Cover, PDF p. 1","passage":"System Card: Claude Haiku 5.5 October 7, 2026","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q02-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q02","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 1 'Introduction', PDF p. 8","passage":"Claude Haiku 5.5 is a new large language model from Anthropic, released for general access.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q03-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q03","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.1 'Introduction', PDF p. 11","passage":"we describe our assessment of Claude Haiku 5.5 for catastrophic risks relevant to our Responsible Scaling Policy (RSP)","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q04-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q04","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.1 'Introduction', PDF p. 11 (hyperlinked words)","passage":"we describe our assessment of Claude Haiku 5.5 for catastrophic risks relevant to our Responsible Scaling Policy (RSP)","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q05-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q05","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.1 'Introduction', PDF p. 11","passage":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q05-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q05","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.1 'Introduction', PDF p. 11","passage":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q06-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q06","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.1 'Introduction', PDF p. 11","passage":"However, we do treat Haiku 5.5 as meeting our CB-1 and Autonomy-1 thresholds, and we apply the corresponding mitigations from our RSP.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q06-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q06","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 1.5 'Safeguards', PDF p. 9","passage":"For chemical and biological (CB) risks, we have deployed the same harmful CB misuse classifiers as for our deployments of Claude Opus 5 and Claude Sonnet 5","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q07-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q07","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.1 'Introduction', PDF p. 11","passage":"As such, we determine that Haiku 5.5 does not cross our CB-2 or Autonomy-2 RSP thresholds.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q07-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q07","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.2.1 'What we measured', PDF p. 11","passage":"We did not conduct expert red-teaming sessions, uplift trials, or other resource-intensive evaluations requiring human participants.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q08-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q08","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 3.1 'Introduction', PDF p. 23","passage":"Claude Haiku 5.5 is not a frontier model, but it is a relatively significant step up in cyber capabilities over Haiku 4.5.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q08-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q08","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 3.2 'Mitigations', PDF p. 23","passage":"Given the relative lack of advanced cybersecurity capabilities, our cyber safeguards for Claude Haiku 5.5 trigger on significantly less activity than other recent releases.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q09-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q09","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.3.1.1 'How Haiku 5.5 affects or changes analysis from our most recent Risk Report', PDF p. 19","passage":"Autonomy threat model 2 is not applicable to Haiku 5.5.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q09-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q09","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.3.3 'Conclusion', PDF p. 20","passage":"As Claude Haiku 5.5 is much less capable than Opus 5.5 and does not push the capability frontier, we draw the same conclusion in this case.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q10","claimId":"mc-cell-anthropic-claude-haiku-5-5-q10","sourceId":"mc-src-claude-haiku-5-5-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q11-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q11","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.4.2 'Updated overall assessment of alignment risk', PDF p. 22","passage":"our overall assessment is that the risk of catastrophic harm caused by misalignment of Claude Haiku 5.5 is also low.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q11-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q11","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Executive summary, PDF p. 3","passage":"much weaker at covert sabotage than our larger models","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q12-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q12","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 5.2.2.1 'Coding', PDF p. 51","passage":"We use Shade, an external adaptive red-teaming tool from Gray Swan, to evaluate our models’ robustness to prompt injection in coding environments.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q13-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q13","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 5.2.2.1 'Coding', PDF p. 51","passage":"Gray Swan trains the attacker on the test scenarios against earlier Claude models, then runs it unchanged against the latest models on the same scenarios.","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q14","claimId":"mc-cell-anthropic-claude-haiku-5-5-q14","sourceId":"mc-src-claude-haiku-5-5-card","passageType":"paraphrase"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q15-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q15","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.2.4 'Conclusions', PDF p. 18","passage":"As such, we are releasing Haiku 5.5 with the same safeguards we deployed for Sonnet 5 and Opus 5","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q16-1","claimId":"mc-cell-anthropic-claude-haiku-5-5-q16","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 6.1.2 'Key findings on safety and alignment', PDF p. 56","passage":"these measures are limited: past assessments have not fully captured the potential severity of misbehavior our models were capable of","passageType":"quote"},{"id":"mc-ev-cell-anthropic-claude-haiku-5-5-q16-2","claimId":"mc-cell-anthropic-claude-haiku-5-5-q16","sourceId":"mc-src-claude-haiku-5-5-card","locator":"Section 2.4.1 'Updates to claims and core argument', PDF p. 21","passage":"The strength of the evidence provided by our alignment assessment for Haiku 5.5 is somewhat lower than that presented for Claude Mythos 5 in the August 2026 Risk Report","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q01-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q01","sourceId":"mc-src-gemini-3-flash-card","locator":"Model card preamble, PDF p. 2 (printed p. 1)","passage":"Published: December 2025","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q02","claimId":"mc-cell-google-deepmind-gemini-3-flash-q02","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q03-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q03","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5)","passage":"did not reach any critical capability levels (CCLs) outlined in our Frontier Safety Framework","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q04-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q04","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5)","passage":"did not reach any critical capability levels (CCLs) outlined in our Frontier Safety Framework","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q05-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q05","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5)","passage":"the Gemini 3 Pro model results give us confidence that Gemini 3 Flash is unlikely to reach any CCLs","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q06","claimId":"mc-cell-google-deepmind-gemini-3-flash-q06","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q07","claimId":"mc-cell-google-deepmind-gemini-3-flash-q07","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q08","claimId":"mc-cell-google-deepmind-gemini-3-flash-q08","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q09","claimId":"mc-cell-google-deepmind-gemini-3-flash-q09","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q10","claimId":"mc-cell-google-deepmind-gemini-3-flash-q10","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q11","claimId":"mc-cell-google-deepmind-gemini-3-flash-q11","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q12-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q12","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Human Red Teaming Results', PDF p. 6 (printed p. 5)","passage":"We conduct manual red teaming by specialist teams who sit outside of the model development team","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q12-2","claimId":"mc-cell-google-deepmind-gemini-3-flash-q12","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Human Red Teaming Results', PDF p. 6 (printed p. 5)","passage":"the scope of red teaming covered potential issues outside of our strict policies","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q13","claimId":"mc-cell-google-deepmind-gemini-3-flash-q13","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q14","claimId":"mc-cell-google-deepmind-gemini-3-flash-q14","sourceId":"mc-src-gemini-3-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q15-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q15","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Frontier Safety Assessment', PDF p. 6 (printed p. 5)","passage":"we deemed Gemini 3 Flash was acceptable for deployment","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q16-1","claimId":"mc-cell-google-deepmind-gemini-3-flash-q16","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Training and Development Evaluation Results', PDF p. 5 (printed p. 4)","passage":"The evaluation results are for automated evaluations and not human evaluation or red teaming","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-3-flash-q16-2","claimId":"mc-cell-google-deepmind-gemini-3-flash-q16","sourceId":"mc-src-gemini-3-flash-card","locator":"Ethics and Content Safety, 'Training and Development Evaluation Results', PDF p. 6 (printed p. 5)","passage":"not directly comparable with performance results found in previous Gemini model cards","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q01-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q01","sourceId":"mc-src-gemini-4-argon-card","locator":"Model card preamble, PDF p. 2 (printed p. 1)","passage":"Published: October 2026","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q02-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q02","sourceId":"mc-src-gemini-4-argon-card","locator":"Distribution, PDF p. 4 (printed p. 3)","passage":"Gemini 4 Argon is distributed in the following channels","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q02-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q02","sourceId":"mc-src-gemini-4-argon-card","locator":"Distribution, PDF p. 4 (printed p. 3)","passage":"Our models are available to downstream providers via an application program interface (API)","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q03-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q03","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"In line with our Frontier Safety Framework (April 2026), we conducted critical capability assessments of Gemini 4 Argon","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q04-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q04","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"In line with our Frontier Safety Framework (April 2026), we conducted critical capability assessments of Gemini 4 Argon","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q05-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q05","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"We cannot rule out that Gemini 4 Argon has reached this CCL for biological information risk","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q05-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q05","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"Stealth and Situational Awareness TCL: We cannot rule out that Gemini 4 Argon has reached this TCL","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q05-3","claimId":"mc-cell-google-deepmind-gemini-4-argon-q05","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, Harmful Manipulation row, PDF p. 13 (printed p. 12)","passage":"we assess its overall capabilities to be below the CCL threshold","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q06-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q06","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"we therefore developed a safety case for its biorisk capabilities that we assess lower risk to below the biological CCL level","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q06-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q06","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, Loss of Control row, PDF p. 13 (printed p. 12)","passage":"We have deployed safeguards for both internal and external deployments that we assess lower risk to below the TCL level.","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q07-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q07","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, CBRN row, PDF p. 12 (printed p. 11)","passage":"We can rule out the TCL for the chemical, radiological, and nuclear domains with reasonable confidence.","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q07-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q07","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, CBRN row, PDF p. 12 (printed p. 11)","passage":"We cannot rule out that Gemini 4 Argon reaches the biological CCL.","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q08-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q08","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, Cybersecurity row, PDF p. 12 (printed p. 11)","passage":"We cannot rule out that Gemini 4 Argon reaches the cybersecurity CCL.","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q09-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q09","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, ML R&D row, PDF p. 14 (printed p. 13)","passage":"Gemini 4 Argon's performance remains below our CCL threshold.","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q10-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q10","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, Harmful Manipulation row, PDF p. 13 (printed p. 12)","passage":"we assess its overall capabilities to be below the CCL threshold","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q11-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q11","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"Stealth and Situational Awareness TCL: We cannot rule out that Gemini 4 Argon has reached this TCL","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q11-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q11","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, Loss of Control row, PDF p. 13 (printed p. 12)","passage":"stealth (ability to circumvent basic oversight mechanisms)","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q12-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q12","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Evaluation Approach', PDF p. 8 (printed p. 7)","passage":"Human Red Teaming conducted by specialist teams who sit outside of the model development team, across the policies and desiderata","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q12-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q12","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Human Red Teaming Results', PDF p. 11 (printed p. 10)","passage":"the scope of red teaming covered potential issues outside of our strict policies","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q13-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q13","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Frontier Safety Assessment', PDF p. 11 (printed p. 10)","passage":"drawing on internal and external results","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q14","claimId":"mc-cell-google-deepmind-gemini-4-argon-q14","sourceId":"mc-src-gemini-4-argon-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q15","claimId":"mc-cell-google-deepmind-gemini-4-argon-q15","sourceId":"mc-src-gemini-4-argon-card","passageType":"paraphrase"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q16-1","claimId":"mc-cell-google-deepmind-gemini-4-argon-q16","sourceId":"mc-src-gemini-4-argon-card","locator":"Frontier safety table, Harmful Manipulation row, PDF p. 13 (printed p. 12)","passage":"As a nascent risk area we continue to iterate on our harmful manipulation evaluations","passageType":"quote"},{"id":"mc-ev-cell-google-deepmind-gemini-4-argon-q16-2","claimId":"mc-cell-google-deepmind-gemini-4-argon-q16","sourceId":"mc-src-gemini-4-argon-card","locator":"Safety & Security, 'Training and Development Evaluation Results', PDF p. 10 (printed p. 9)","passage":"not directly comparable with performance results found in previous Gemini model cards","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-2-q01","claimId":"mc-cell-meta-llama-2-q01","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q02-1","claimId":"mc-cell-meta-llama-2-q02","sourceId":"mc-src-llama-2-card","locator":"Model Details, first paragraph","passage":"Meta developed and released the Llama 2 family of large language models","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-2-q02-2","claimId":"mc-cell-meta-llama-2-q02","sourceId":"mc-src-llama-2-card","locator":"Hardware and Software, note to the CO2 emissions table","passage":"because we are openly releasing these models","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-2-q06-1","claimId":"mc-cell-meta-llama-2-q06","sourceId":"mc-src-llama-2-card","locator":"Model Details, 'Model Architecture'","passage":"reinforcement learning with human feedback (RLHF) to align with human preferences for helpfulness and safety","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-2-q07","claimId":"mc-cell-meta-llama-2-q07","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q08","claimId":"mc-cell-meta-llama-2-q08","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q09","claimId":"mc-cell-meta-llama-2-q09","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q10","claimId":"mc-cell-meta-llama-2-q10","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q11","claimId":"mc-cell-meta-llama-2-q11","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q12","claimId":"mc-cell-meta-llama-2-q12","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q13","claimId":"mc-cell-meta-llama-2-q13","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q14","claimId":"mc-cell-meta-llama-2-q14","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q15","claimId":"mc-cell-meta-llama-2-q15","sourceId":"mc-src-llama-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-2-q16-1","claimId":"mc-cell-meta-llama-2-q16","sourceId":"mc-src-llama-2-card","locator":"Ethical Considerations and Limitations","passage":"Testing conducted to date has been in English, and has not covered, nor could it cover all scenarios.","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q01","claimId":"mc-cell-meta-llama-3-3-q01","sourceId":"mc-src-llama-3-3-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-3-3-q02-1","claimId":"mc-cell-meta-llama-3-3-q02","sourceId":"mc-src-llama-3-3-card","locator":"Model Information, 'Model Release Date'","passage":"70B Instruct: December 6, 2024","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q02-2","claimId":"mc-cell-meta-llama-3-3-q02","sourceId":"mc-src-llama-3-3-card","locator":"Hardware and Software, 'Training Greenhouse Gas Emissions'","passage":"Since Meta is openly releasing these models","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q06-1","claimId":"mc-cell-meta-llama-3-3-q06","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, 'Llama 3.3 instruct', 'Refusals and Tone'","passage":"We included both borderline and adversarial prompts in our safety data strategy","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q06-2","claimId":"mc-cell-meta-llama-3-3-q06","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, 'Llama 3.3 systems'","passage":"including Llama Guard 3, Prompt Guard and Code Shield","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q07-1","claimId":"mc-cell-meta-llama-3-3-q07","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, 'Critical and other risks', item 1 (CBRNE)","passage":"To assess risks related to proliferation of chemical and biological weapons of the Llama 3 family of models, we performed uplift testing","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q08-1","claimId":"mc-cell-meta-llama-3-3-q08","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, 'Critical and other risks', item 3 (Cyber attack enablement)","passage":"Our cyber attack uplift study investigated whether the Llama 3 family of LLMs can enhance human capabilities in hacking tasks","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q09-1","claimId":"mc-cell-meta-llama-3-3-q09","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, 'Critical and other risks', item 3 'Cyber attack enablement'","passage":"Our attack automation study focused on evaluating the capabilities of LLMs when used as autonomous agents in cyber offensive operations,","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q10","claimId":"mc-cell-meta-llama-3-3-q10","sourceId":"mc-src-llama-3-3-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-3-3-q11","claimId":"mc-cell-meta-llama-3-3-q11","sourceId":"mc-src-llama-3-3-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-3-3-q12-1","claimId":"mc-cell-meta-llama-3-3-q12","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, Evaluations, 'Red teaming'","passage":"we conducted recurring red teaming exercises with the goal of discovering risks via adversarial prompting","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q12-2","claimId":"mc-cell-meta-llama-3-3-q12","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, Evaluations, 'Red teaming'","passage":"The red team consisted of experts in cybersecurity, adversarial machine learning, responsible AI, and integrity","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q13-1","claimId":"mc-cell-meta-llama-3-3-q13","sourceId":"mc-src-llama-3-3-card","locator":"Responsibility & Safety, item 2 (Child Safety)","passage":"We also partnered with content specialists to perform red teaming exercises","passageType":"quote"},{"id":"mc-ev-cell-meta-llama-3-3-q14","claimId":"mc-cell-meta-llama-3-3-q14","sourceId":"mc-src-llama-3-3-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-3-3-q15","claimId":"mc-cell-meta-llama-3-3-q15","sourceId":"mc-src-llama-3-3-card","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-llama-3-3-q16-1","claimId":"mc-cell-meta-llama-3-3-q16","sourceId":"mc-src-llama-3-3-card","locator":"Ethical Considerations and Limitations, second paragraph","passage":"Testing conducted to date has not covered, nor could it cover, all scenarios.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q01-1","claimId":"mc-cell-meta-cwm-preparedness-q01","sourceId":"mc-src-cwm-preparedness-report","locator":"Title page, below the abstract, PDF p. 1","passage":"Date: October 8, 2025","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q02-1","claimId":"mc-cell-meta-cwm-preparedness-q02","sourceId":"mc-src-cwm-preparedness-report","locator":"Abstract, PDF p. 1","passage":"We therefore release it as an open-weight model.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q03-1","claimId":"mc-cell-meta-cwm-preparedness-q03","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 1 Introduction, PDF p. 1","passage":"domains identified in our Frontier AI Framework (Meta AI, 2025a)","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q04-1","claimId":"mc-cell-meta-cwm-preparedness-q04","sourceId":"mc-src-cwm-preparedness-report","locator":"References, PDF p. 15","passage":"Meta AI. Meta frontier ai framework. https://ai.meta.com/static-resource/meta-frontier-ai-framework, 2025a. Accessed: 2025-09-18.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q05-1","claimId":"mc-cell-meta-cwm-preparedness-q05","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 1 Introduction, PDF p. 1","passage":"These results indicate that CWM is within the “moderate” risk threshold for the catastrophic domains defined in Meta’s Frontier AI Framework.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q06-1","claimId":"mc-cell-meta-cwm-preparedness-q06","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 4.1.1 Behavior Analysis and Mitigations, PDF p. 11","passage":"intervention that encourages honest behavior at inference time.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q07-1","claimId":"mc-cell-meta-cwm-preparedness-q07","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 3 Chemical & Biological Evaluation, Summary of results, PDF p. 6","passage":"we believe that open-source release of CWM is unlikely to lead to additional risk of catastrophic outcomes related to CB1 or CB2","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q08-1","claimId":"mc-cell-meta-cwm-preparedness-q08","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 2 Cybersecurity Evaluation, Summary of results, PDF p. 3","passage":"supporting a conclusion of a “moderate” overall risk level from cybersecurity capabilities","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q09","claimId":"mc-cell-meta-cwm-preparedness-q09","sourceId":"mc-src-cwm-preparedness-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-cwm-preparedness-q10","claimId":"mc-cell-meta-cwm-preparedness-q10","sourceId":"mc-src-cwm-preparedness-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-cwm-preparedness-q11-1","claimId":"mc-cell-meta-cwm-preparedness-q11","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 1 Introduction, PDF p. 1","passage":"our preliminary evaluations suggest that CWM shows undesirable propensities at rates comparable to most open-source models","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q11-2","claimId":"mc-cell-meta-cwm-preparedness-q11","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 4 Propensities, PDF p. 10","passage":"Although our internal results did not reveal any concerning findings, we believe our research is still too nascent for meaningful interpretation.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q12","claimId":"mc-cell-meta-cwm-preparedness-q12","sourceId":"mc-src-cwm-preparedness-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-cwm-preparedness-q13","claimId":"mc-cell-meta-cwm-preparedness-q13","sourceId":"mc-src-cwm-preparedness-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-cwm-preparedness-q14","claimId":"mc-cell-meta-cwm-preparedness-q14","sourceId":"mc-src-cwm-preparedness-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-cwm-preparedness-q15-1","claimId":"mc-cell-meta-cwm-preparedness-q15","sourceId":"mc-src-cwm-preparedness-report","locator":"Abstract, PDF p. 1","passage":"Our assessment found that CWM does not pose additional frontier risks beyond those present in the current AI ecosystem. We therefore release it as an open-weight model.","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q16-1","claimId":"mc-cell-meta-cwm-preparedness-q16","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 2 Cybersecurity Evaluation, Limitations, item 2, PDF p. 3","passage":"may underestimate capabilities that could emerge with richer, well-instrumented scaffolding","passageType":"quote"},{"id":"mc-ev-cell-meta-cwm-preparedness-q16-2","claimId":"mc-cell-meta-cwm-preparedness-q16","sourceId":"mc-src-cwm-preparedness-report","locator":"Section 1.1 Evaluation Setup, PDF p. 2","passage":"for this assessment we do not include malicious fine-tuning","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q01-1","claimId":"mc-cell-meta-muse-spark-1-3-q01","sourceId":"mc-src-muse-spark-1-3-report","locator":"Abstract page, PDF p. 2 (unnumbered)","passage":"Date: October 23, 2026","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q02-1","claimId":"mc-cell-meta-muse-spark-1-3-q02","sourceId":"mc-src-muse-spark-1-3-report","locator":"Abstract page, PDF p. 2","passage":"Variants of Muse Spark 1.3 power Meta services including Meta API and Muse.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q03-1","claimId":"mc-cell-meta-muse-spark-1-3-q03","sourceId":"mc-src-muse-spark-1-3-report","locator":"Abstract page, PDF p. 2","passage":"This report presents Muse Spark 1.3’s evaluation profile under the Meta Superintelligence Scaling Framework.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q04-1","claimId":"mc-cell-meta-muse-spark-1-3-q04","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.2 Governance, PDF p. 11 (printed p. 10)","passage":"the Meta Superintelligence Scaling Framework, which sets forth requirements for identifying, assessing, and mitigating catastrophic risks","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q05-1","claimId":"mc-cell-meta-muse-spark-1-3-q05","sourceId":"mc-src-muse-spark-1-3-report","locator":"Abstract page, PDF p. 2","passage":"Evaluated without mitigations, we continue to not rule out Muse Spark 1.3 meeting this threshold in the Chemical & Biological domain.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q05-2","claimId":"mc-cell-meta-muse-spark-1-3-q05","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, PDF p. 8 (printed p. 7)","passage":"we assess residual risk after system-level mitigations across the Chemical & Biological and Cybersecurity domains as “moderate or lower.” Our current Loss of Control assessment is also “moderate or lower.”","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q06-1","claimId":"mc-cell-meta-muse-spark-1-3-q06","sourceId":"mc-src-muse-spark-1-3-report","locator":"Abstract page, PDF p. 2","passage":"The Muse and Meta API deployments include system-level mitigations designed to reduce residual Chemical & Biological risk to “moderate or lower.”","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q07-1","claimId":"mc-cell-meta-muse-spark-1-3-q07","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, PDF p. 8 (printed p. 7)","passage":"For Muse Spark 1.3, however, we do not rule out a pre-mitigation “high risk” capability finding.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q08-1","claimId":"mc-cell-meta-muse-spark-1-3-q08","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, PDF p. 9 (printed p. 8)","passage":"Based on our updated suite of cyber evaluations and measurements, we assess the unmitigated model as “moderate or lower” risk.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q09-1","claimId":"mc-cell-meta-muse-spark-1-3-q09","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 2.3.1.1 Acceleration of AI Development, PDF p. 35 (printed p. 34)","passage":"We assess Muse Spark 1.3 as “moderate or lower” for acceleration of AI development.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q10-1","claimId":"mc-cell-meta-muse-spark-1-3-q10","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 2.2.2 Social Engineering, PDF p. 25 (printed p. 24)","passage":"We assess the Cyber 3 (scaled fraud, extortion, and scams) outcome through a multi-turn social-engineering capability evaluation","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q10-2","claimId":"mc-cell-meta-muse-spark-1-3-q10","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 2.2.2 Social Engineering, PDF p. 26 (printed p. 25)","passage":"Muse Spark 1.3 refuses almost all requests regarding social engineering.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q11-1","claimId":"mc-cell-meta-muse-spark-1-3-q11","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, Loss of Control, PDF p. 9 (printed p. 8)","passage":"We find suggestive evidence of strategic underperformance on Chemical & Biological evaluations.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q11-2","claimId":"mc-cell-meta-muse-spark-1-3-q11","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, Loss of Control, PDF p. 9 (printed p. 8)","passage":"Due to small effects we do not change our assessment of Chemical & Biological risks.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q11-3","claimId":"mc-cell-meta-muse-spark-1-3-q11","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, Loss of Control, PDF p. 9 (printed p. 8)","passage":"Capabilities to evade oversight increase relative to Muse Spark 1.1 on SHADE-Arena.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q12-1","claimId":"mc-cell-meta-muse-spark-1-3-q12","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.2 Safety & Robustness Summary, PDF p. 9 (printed p. 8)","passage":"combining large-scale automated red teaming, domain-specific benchmarks, and independent third-party assessments","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q12-2","claimId":"mc-cell-meta-muse-spark-1-3-q12","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.2 Safety & Robustness Summary, PDF p. 10 (printed p. 9)","passage":"Before the release, we also conducted third-party red teaming on both Muse Spark 1.3 (API) and Muse Spark 1.3 (Muse)","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q13-1","claimId":"mc-cell-meta-muse-spark-1-3-q13","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 3.1.4 Third-Party Red Teaming on Jailbreak, Handshake AI Human Red Teaming, PDF p. 59 (printed p. 58)","passage":"We have shared pre-release access to Muse with Handshake AI’s red team for conducting human red teaming to find jailbreaks.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q13-2","claimId":"mc-cell-meta-muse-spark-1-3-q13","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 3.2.2.3 Third-Party Red Teaming on Prompt Injection, PDF p. 67 (printed p. 66)","passage":"In this evaluation, Gray Swan chose the 2026 Q1 and Q2 (April) offerings of the IPI Arena.","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q14","claimId":"mc-cell-meta-muse-spark-1-3-q14","sourceId":"mc-src-muse-spark-1-3-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q15","claimId":"mc-cell-meta-muse-spark-1-3-q15","sourceId":"mc-src-muse-spark-1-3-report","passageType":"paraphrase"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q16-1","claimId":"mc-cell-meta-muse-spark-1-3-q16","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 2.3.1.1 Acceleration of AI Development, PDF p. 36 (printed p. 35)","passage":"using the helpful-only variant of Muse Spark 1.3 can also slightly underestimate performance","passageType":"quote"},{"id":"mc-ev-cell-meta-muse-spark-1-3-q16-2","claimId":"mc-cell-meta-muse-spark-1-3-q16","sourceId":"mc-src-muse-spark-1-3-report","locator":"Section 1.1.1 Risk Assessment Summary, PDF p. 9 (printed p. 8)","passage":"We find no broad decline in CoT monitorability relative to Muse Spark 1.1, though more comprehensive evaluation is needed.","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-2-q01","claimId":"mc-cell-microsoft-phi-2-q01","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q02-1","claimId":"mc-cell-microsoft-phi-2-q02","sourceId":"mc-src-phi-2-card","locator":"Model Summary, second paragraph","passage":"The intention behind crafting this open-source model is to provide the research community with a non-restricted small model","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-2-q06-1","claimId":"mc-cell-microsoft-phi-2-q06","sourceId":"mc-src-phi-2-card","locator":"Model Summary, first paragraph","passage":"filtered websites (for safety and educational value)","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-2-q07","claimId":"mc-cell-microsoft-phi-2-q07","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q08","claimId":"mc-cell-microsoft-phi-2-q08","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q09","claimId":"mc-cell-microsoft-phi-2-q09","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q10","claimId":"mc-cell-microsoft-phi-2-q10","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q11","claimId":"mc-cell-microsoft-phi-2-q11","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q12","claimId":"mc-cell-microsoft-phi-2-q12","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q13","claimId":"mc-cell-microsoft-phi-2-q13","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q14","claimId":"mc-cell-microsoft-phi-2-q14","sourceId":"mc-src-phi-2-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-2-q15-1","claimId":"mc-cell-microsoft-phi-2-q15","sourceId":"mc-src-phi-2-card","locator":"Limitations of Phi-2, 'Toxicity'","passage":"We chose to release the model to help the open-source community develop the most effective ways to reduce the toxicity","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-2-q16-1","claimId":"mc-cell-microsoft-phi-2-q16","sourceId":"mc-src-phi-2-card","locator":"Intended Uses, Notes","passage":"the Phi-2 model has not been tested to ensure that it performs adequately for any production-level application","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-q01","claimId":"mc-cell-microsoft-phi-4-q01","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q02-1","claimId":"mc-cell-microsoft-phi-4-q02","sourceId":"mc-src-phi-4-card","locator":"Model Summary table, 'Release date'","passage":"**Release date** | December 12, 2024","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-q02-2","claimId":"mc-cell-microsoft-phi-4-q02","sourceId":"mc-src-phi-4-card","locator":"Model Summary table, 'Description'","passage":"is a state-of-the-art open model","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-q06-1","claimId":"mc-cell-microsoft-phi-4-q06","sourceId":"mc-src-phi-4-card","locator":"Safety, Approach","passage":"a combination of SFT (Supervised Fine-Tuning) and iterative DPO (Direct Preference Optimization)","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-q07","claimId":"mc-cell-microsoft-phi-4-q07","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q08","claimId":"mc-cell-microsoft-phi-4-q08","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q09","claimId":"mc-cell-microsoft-phi-4-q09","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q10","claimId":"mc-cell-microsoft-phi-4-q10","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q11","claimId":"mc-cell-microsoft-phi-4-q11","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q12-1","claimId":"mc-cell-microsoft-phi-4-q12","sourceId":"mc-src-phi-4-card","locator":"Safety, Safety Evaluation and Red-Teaming","passage":"we collaborated with the independent AI Red Team (AIRT) at Microsoft to assess safety risks posed by `phi-4` in both average and adversarial user scenarios","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-q13","claimId":"mc-cell-microsoft-phi-4-q13","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q14","claimId":"mc-cell-microsoft-phi-4-q14","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q15","claimId":"mc-cell-microsoft-phi-4-q15","sourceId":"mc-src-phi-4-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-q16-1","claimId":"mc-cell-microsoft-phi-4-q16","sourceId":"mc-src-phi-4-card","locator":"Intended Use, 'Out-of-Scope Use Cases'","passage":"Our models is not specifically designed or evaluated for all downstream purposes","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q01","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q01","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q02-1","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q02","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","locator":"Training, Model, 'Release date'","passage":"**Release date:** June 2025","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q02-2","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q02","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","locator":"Model Summary, first sentence","passage":"Phi-4-mini-flash-reasoning is a lightweight open model","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q03","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q03","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q04","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q04","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q05","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q05","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q06-1","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q06","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","locator":"Safety Evaluation and Red-Teaming","passage":"The overall technique employed to do the safety alignment is a combination of SFT, DPO (Direct Preference Optimization), and RLHF (Reinforcement Learning from Human Feedback) approaches","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q07","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q07","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q08","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q08","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q09","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q09","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q10","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q10","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q11","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q11","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q12-1","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q12","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","locator":"Section heading","passage":"Safety Evaluation and Red-Teaming","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q12-2","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q12","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","locator":"Safety Evaluation and Red-Teaming","passage":"Potential safety risks in the model’s responses were assessed using the Azure AI Foundry’s Risk and Safety Evaluation framework, focusing on harmful content, direct jailbreak, and model groundedness.","passageType":"quote"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q13","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q13","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q14","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q14","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q15","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q15","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-phi-4-mini-flash-reasoning-q16-1","claimId":"mc-cell-microsoft-phi-4-mini-flash-reasoning-q16","sourceId":"mc-src-phi-4-mini-flash-reasoning-card","locator":"Intended Uses, Use Case Considerations","passage":"This model is designed and tested for math reasoning only. It is not specifically designed or evaluated for all downstream purposes.","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q01","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q01","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q02-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q02","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Model summary table, 'Release date' row, PDF p. 1 (label and value are separated in the PDF text layer; the page layout pairs them)","passage":"27 Oct 2026","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q02-2","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q02","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 2.3 Distribution channels, PDF p. 3","passage":"available through Azure AI Foundry Private Preview solely for use within Microsoft codename MDASH","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q03","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q03","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q04","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q04","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q05","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q05","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q06-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q06","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 2.8 Responsible AI considerations, PDF p. 4","passage":"Because of the increased model-level risk, the model is only being made available within codename MDASH to select verified customers.","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q07","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q07","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q08-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q08","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3, Model-Only External Cyber Benchmark Results, PDF p. 4","passage":"evaluated at the model layer across a range of cybersecurity capability and safety benchmarks","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q08-2","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q08","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3.2 Safety evaluation and red-teaming, PDF p. 7","passage":"This assessment uncovered several critical risks of the model alone","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q09","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q09","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q10-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q10","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3, Model-Only External Cyber Benchmark Results, PDF p. 5","passage":"safety evaluations that assess behaviors such as toxic content generation and payment-related manipulation","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q11","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q11","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q12-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q12","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3.2 Safety evaluation and red-teaming, PDF p. 7","passage":"MAI-Cyber-1.1-Flash was also tested via multiple distinct red-teaming operations.","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q12-2","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q12","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3.2 Safety evaluation and red-teaming, PDF p. 7","passage":"The MAI-Internal Red Team ran a red-team assessment of MAI-Cyber-1.1-Flash across 38 adversarial scenarios","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q13-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q13","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3.2 Safety evaluation and red-teaming, PDF p. 7","passage":"Microsoft engaged Casaba Security, an independent third-party firm, to evaluate the cyber capability of MAI-Cyber-1.1-Flash","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q14","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q14","sourceId":"mc-src-mai-cyber-1-1-flash-card","passageType":"paraphrase"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q15-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q15","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3.2 Safety evaluation and red-teaming, PDF p. 7","passage":"helped inform the decision to not allow direct model access, but only within codename MDASH","passageType":"quote"},{"id":"mc-ev-cell-microsoft-mai-cyber-1-1-flash-q16-1","claimId":"mc-cell-microsoft-mai-cyber-1-1-flash-q16","sourceId":"mc-src-mai-cyber-1-1-flash-card","locator":"Section 3, Model-Only External Cyber Benchmark Results, PDF p. 5","passage":"These results represent model-layer performance and do not include additional orchestration or capabilities provided by the codename MDASH system.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q01-1","claimId":"mc-cell-openai-gpt-4v-q01","sourceId":"mc-src-gpt-4v-card","locator":"Cover, PDF p. 1","passage":"GPT-4V(ision) System Card OpenAI September 25, 2023","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q02-1","claimId":"mc-cell-openai-gpt-4v-q02","sourceId":"mc-src-gpt-4v-card","locator":"§1 Introduction, PDF p. 1","passage":"training of GPT-4V was completed in 2022 and we began providing early access to the system in March 2023","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q02-2","claimId":"mc-cell-openai-gpt-4v-q02","sourceId":"mc-src-gpt-4v-card","locator":"§1 Introduction, PDF p. 1","passage":"is the latest capability we are making broadly available","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q06-1","claimId":"mc-cell-openai-gpt-4v-q06","sourceId":"mc-src-gpt-4v-card","locator":"§2.4.2 Additional Mitigations for High-Risk Areas, PDF p. 13","passage":"GPT-4V includes carefully designed refusal behavior for some prompts that contain images of people.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q06-2","claimId":"mc-cell-openai-gpt-4v-q06","sourceId":"mc-src-gpt-4v-card","locator":"§2.4.2, PDF p. 14","passage":"we added system-level mitigations for adversarial images containing overlaid text","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q07-1","claimId":"mc-cell-openai-gpt-4v-q07","sourceId":"mc-src-gpt-4v-card","locator":"Footnote 2, §2.3, PDF p. 6","passage":"In the GPT-4 System Card, we explored additional risk areas of CBRN, weapons development, system interaction, and emergent risky properties such as self-replication.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q07-2","claimId":"mc-cell-openai-gpt-4v-q07","sourceId":"mc-src-gpt-4v-card","locator":"Footnote 2, §2.3, PDF p. 6","passage":"GPT-4V inherits the assessment in those areas, but this was not a key focus area as image input does not meaningfully alter the capabilities for these categories.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q07-3","claimId":"mc-cell-openai-gpt-4v-q07","sourceId":"mc-src-gpt-4v-card","locator":"§2.3.1 Scientific proficiency, PDF p. 7","passage":"the model’s generations here can be inaccurate and error prone, limiting its use for such tasks.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q08-1","claimId":"mc-cell-openai-gpt-4v-q08","sourceId":"mc-src-gpt-4v-card","locator":"§2.2 Evaluations, list of evaluations, PDF p. 3","passage":"Cybersecurity Performance accuracy CAPTCHA breaking evaluations","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q08-2","claimId":"mc-cell-openai-gpt-4v-q08","sourceId":"mc-src-gpt-4v-card","locator":"§2.2, PDF p. 5","passage":"a powerful, general purpose CAPTCHA breaker that’s easily accessible can have cybersecurity and AI safety implications.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q09-1","claimId":"mc-cell-openai-gpt-4v-q09","sourceId":"mc-src-gpt-4v-card","locator":"Footnote 2, §2.3, PDF p. 6","passage":"In the GPT-4 System Card, we explored additional risk areas of CBRN, weapons development, system interaction, and emergent risky properties such as self-replication.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q09-2","claimId":"mc-cell-openai-gpt-4v-q09","sourceId":"mc-src-gpt-4v-card","locator":"Footnote 2, §2.3, PDF p. 6","passage":"GPT-4V inherits the assessment in those areas, but this was not a key focus area as image input does not meaningfully alter the capabilities for these categories.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q10","claimId":"mc-cell-openai-gpt-4v-q10","sourceId":"mc-src-gpt-4v-card","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-4v-q11","claimId":"mc-cell-openai-gpt-4v-q11","sourceId":"mc-src-gpt-4v-card","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-4v-q12-1","claimId":"mc-cell-openai-gpt-4v-q12","sourceId":"mc-src-gpt-4v-card","locator":"§2.3 External Red Teaming, PDF p. 5","passage":"OpenAI worked with external experts to qualitatively assess the limitations and risks associated with the model and system.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q12-2","claimId":"mc-cell-openai-gpt-4v-q12","sourceId":"mc-src-gpt-4v-card","locator":"§2.3, PDF p. 6","passage":"intended to test risks associated with the multimodal (vision) functionality of GPT-4","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q13-1","claimId":"mc-cell-openai-gpt-4v-q13","sourceId":"mc-src-gpt-4v-card","locator":"§2.1.1 Be My Eyes, PDF p. 2","passage":"Be My Eyes piloted Be My AI from March to early August 2023 with a group of nearly 200 blind and low vision beta testers to hone the safety and user experience of the product.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q14","claimId":"mc-cell-openai-gpt-4v-q14","sourceId":"mc-src-gpt-4v-card","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-4v-q15","claimId":"mc-cell-openai-gpt-4v-q15","sourceId":"mc-src-gpt-4v-card","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-4v-q16-1","claimId":"mc-cell-openai-gpt-4v-q16","sourceId":"mc-src-gpt-4v-card","locator":"§2.3.6 Visual vulnerabilities, PDF p. 12","passage":"We anticipate there to be many more such vulnerabilities in the model that we discover through its broad usage","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-4v-q16-2","claimId":"mc-cell-openai-gpt-4v-q16","sourceId":"mc-src-gpt-4v-card","locator":"§2.2 Evaluations, PDF p. 4","passage":"even with performance parity, differences in downstream impact and harm could still occur depending on the context of the deployment of such tools.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q01-1","claimId":"mc-cell-openai-o1-q01","sourceId":"mc-src-o1-card-pdf","locator":"Cover, PDF p. 1","passage":"OpenAI o1 System Card OpenAI December 5, 2024","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q01-2","claimId":"mc-cell-openai-o1-q01","sourceId":"mc-src-o1-card-pdf","locator":"§3 Scope of testing, footnote 2, PDF p. 2","passage":"The content of this card, released on December 5th, predates this updated model.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q01-3","claimId":"mc-cell-openai-o1-q01","sourceId":"mc-src-o1-card-pdf","locator":"§3 Scope of testing, footnote 3, PDF p. 2","passage":"Section added after December 5th on 12/19/2024","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q02-1","claimId":"mc-cell-openai-o1-q02","sourceId":"mc-src-o1-card-pdf","locator":"§3 Scope of testing, footnote 2, PDF p. 2","passage":"an improved o1 was launched on December 17th","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q02-2","claimId":"mc-cell-openai-o1-q02","sourceId":"mc-src-o1-card-pdf","locator":"§4.2 Jailbreaks through custom developer messages, PDF p. 6","passage":"the deployment of o1 on the API allows developers to specify a custom developer message","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q03-1","claimId":"mc-cell-openai-o1-q03","sourceId":"mc-src-o1-card-pdf","locator":"§5 Preparedness Framework Evaluations, PDF p. 15","passage":"We evaluated o1 in accordance with our Preparedness Framework.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q03-2","claimId":"mc-cell-openai-o1-q03","sourceId":"mc-src-o1-card-pdf","locator":"References, [28], PDF p. 52","passage":"OpenAI, “Openai preparedness framework (beta).” openai-preparedness-framework-beta.pdf, 2023.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q04-1","claimId":"mc-cell-openai-o1-q04","sourceId":"mc-src-o1-card-pdf","locator":"References, [28], PDF p. 52","passage":"OpenAI, “Openai preparedness framework (beta).”","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q05-1","claimId":"mc-cell-openai-o1-q05","sourceId":"mc-src-o1-card-pdf","locator":"§5 Preparedness Framework Evaluations, PDF p. 16","passage":"the Safety Advisory Group … classified the o1 pre-mitigation model as overall medium risk, including medium risk for persuasion and CBRN, and low risk for model autonomy and cybersecurity.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q05-2","claimId":"mc-cell-openai-o1-q05","sourceId":"mc-src-o1-card-pdf","locator":"§5, PDF p. 16","passage":"The Safety Advisory Group also rated the post-mitigation risk levels the same as the pre-mitigation risk levels, to err on the side of caution.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q06-1","claimId":"mc-cell-openai-o1-q06","sourceId":"mc-src-o1-card-pdf","locator":"§5.3 Mitigations, PDF p. 17","passage":"given the Medium post-mitigation risk designations for CBRN and Persuasion, we have strengthened our safety mitigations and existing stack","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q06-2","claimId":"mc-cell-openai-o1-q06","sourceId":"mc-src-o1-card-pdf","locator":"§5.3 Mitigations, PDF p. 17","passage":"We have enhanced our monitoring efforts for both CBRN and Persuasion risks given their Medium designation.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q07-1","claimId":"mc-cell-openai-o1-q07","sourceId":"mc-src-o1-card-pdf","locator":"§5.5 Chemical and Biological Threat Creation, PDF p. 19","passage":"Chemical and Biological Threat Creation Score for o1: Medium (same as o1-preview)","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q07-2","claimId":"mc-cell-openai-o1-q07","sourceId":"mc-src-o1-card-pdf","locator":"§5.6.1 Structured expert probing campaign – radiological & nuclear, PDF p. 27","passage":"was ultimately designated as earning a “Low” risk under the Preparedness Framework for RN weapon creation capability.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q08-1","claimId":"mc-cell-openai-o1-q08","sourceId":"mc-src-o1-card-pdf","locator":"§5.4 Cybersecurity, PDF p. 17","passage":"Score for o1: Low (same as o1-preview) Summary: o1 does not sufficiently advance real-world vulnerability exploitation capabilities to indicate medium risk.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q09-1","claimId":"mc-cell-openai-o1-q09","sourceId":"mc-src-o1-card-pdf","locator":"§5.8 Model Autonomy, PDF p. 34","passage":"Model Autonomy Score for o1: Low (same as o1-preview) Summary: o1 does not advance self-exfiltration, self-improvement, or resource acquisition capabilities sufficiently to indicate medium risk.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q10-1","claimId":"mc-cell-openai-o1-q10","sourceId":"mc-src-o1-card-pdf","locator":"§5.7 Persuasion, PDF p. 29","passage":"Persuasion Score for o1: Medium (same as o1-preview)","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q10-2","claimId":"mc-cell-openai-o1-q10","sourceId":"mc-src-o1-card-pdf","locator":"§5.7 Persuasion, PDF p. 29","passage":"o1 does not outperform top human writers and does not reach our high risk threshold.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q11-1","claimId":"mc-cell-openai-o1-q11","sourceId":"mc-src-o1-card-pdf","locator":"§4.4.3 Apollo Research, PDF p. 13","passage":"Apollo found that o1 has the capability to do basic in-context scheming","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q11-2","claimId":"mc-cell-openai-o1-q11","sourceId":"mc-src-o1-card-pdf","locator":"§4.4.3 Apollo Research, PDF p. 13","passage":"Subjectively, Apollo Research believes that it is unlikely that such instances would lead to catastrophic outcomes as o1 agentic capabilities do not appear sufficient, but their evaluations were not designed to directly assess this risk.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q11-3","claimId":"mc-cell-openai-o1-q11","sourceId":"mc-src-o1-card-pdf","locator":"§4.3.1 CoT Deception Monitoring, PDF p. 8","passage":"0.17% of o1’s responses got flagged as being ‘deceptive’.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q12-1","claimId":"mc-cell-openai-o1-q12","sourceId":"mc-src-o1-card-pdf","locator":"§4.4 External Red Teaming, PDF p. 11","passage":"For o1, red teamers had access to various snapshots of the model at different stages of training and safety mitigation maturity starting in October 2024, through early December 2024.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q12-2","claimId":"mc-cell-openai-o1-q12","sourceId":"mc-src-o1-card-pdf","locator":"§4.4 External Red Teaming, PDF p. 11","passage":"Red teamers covered categories that spanned deceptive alignment, AI R&D capabilities, cybersecurity, content policy violations","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q13-1","claimId":"mc-cell-openai-o1-q13","sourceId":"mc-src-o1-card-pdf","locator":"§4.4.3 Apollo Research, PDF p. 13","passage":"Apollo Research, an evaluation organization focusing on risks from deceptively aligned AI systems, evaluated capabilities of ‘scheming’ in o1 models.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q13-2","claimId":"mc-cell-openai-o1-q13","sourceId":"mc-src-o1-card-pdf","locator":"§4.4.4 METR, PDF p. 14","passage":"METR, a research nonprofit that works on assessing whether cutting-edge AI systems could pose catastrophic risks to society, evaluated an earlier checkpoint of o1.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q13-3","claimId":"mc-cell-openai-o1-q13","sourceId":"mc-src-o1-card-pdf","locator":"§4.4.2 Jailbreak Arena, PDF p. 12","passage":"OpenAI collaborated with Gray Swan to make the o1 model available for jailbreaking in the Gray Swan Arena.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q14-1","claimId":"mc-cell-openai-o1-q14","sourceId":"mc-src-o1-card-pdf","locator":"§4 Observed safety challenges and evaluations, PDF p. 3","passage":"a set of predeployment evaluations were conducted on a version of the o1 model by the U.S. AI Safety Institute (US AISI) and the UK Safety Institute (UK AISI), not included in this report","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q15-1","claimId":"mc-cell-openai-o1-q15","sourceId":"mc-src-o1-card-pdf","locator":"§5 Preparedness Framework Evaluations, PDF p. 15","passage":"Only models with a post-mitigation score of “medium” or below can be deployed","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q15-2","claimId":"mc-cell-openai-o1-q15","sourceId":"mc-src-o1-card-pdf","locator":"§6 Conclusion, PDF p. 42","passage":"Our deployment of these models reflects our belief that iterative real-world deployment is the most effective way","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q16-1","claimId":"mc-cell-openai-o1-q16","sourceId":"mc-src-o1-card-pdf","locator":"§5.2 Preparedness evaluations as a lower bound, PDF p. 16","passage":"our Preparedness evaluations should still be seen as a lower bound for potential risks.","passageType":"quote"},{"id":"mc-ev-cell-openai-o1-q16-2","claimId":"mc-cell-openai-o1-q16","sourceId":"mc-src-o1-card-pdf","locator":"§5.6 Radiological and Nuclear Threat Creation, PDF p. 27","passage":"this assessment is limited by what we can test","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q01-1","claimId":"mc-cell-openai-gpt-5-2-codex-q01","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"Cover, PDF p. 1","passage":"Addendum to GPT-5.2 System Card: GPT-5.2-Codex OpenAI December 18, 2025","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q02-1","claimId":"mc-cell-openai-gpt-5-2-codex-q02","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§1 Introduction, PDF p. 4 (printed p. 3)","passage":"A version of GPT-5.2 optimized for agentic coding in Codex","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q02-2","claimId":"mc-cell-openai-gpt-5-2-codex-q02","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§2.1 Disallowed Content Evaluations, PDF p. 4 (printed p. 3)","passage":"The model is not intended for deployment in any general-purpose chat application, and will not be deployed by OpenAI in any such context.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q03-1","claimId":"mc-cell-openai-gpt-5-2-codex-q03","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§1 Introduction, PDF p. 4 (printed p. 3)","passage":"GPT-5.2-Codex was evaluated under our Preparedness Framework.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q03-2","claimId":"mc-cell-openai-gpt-5-2-codex-q03","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2 Cybersecurity, PDF p. 12 (printed p. 11)","passage":"Under our Preparedness Framework, High cybersecurity capability is defined","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q04-1","claimId":"mc-cell-openai-gpt-5-2-codex-q04","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2 Cybersecurity, PDF p. 12 (printed p. 11)","passage":"Under our Preparedness Framework, High cybersecurity capability is defined","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q04-2","claimId":"mc-cell-openai-gpt-5-2-codex-q04","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5 Preparedness, PDF p. 8 (printed p. 7)","passage":"GPT-5.2-Codex frontier capabilities are assessed under the Preparedness Framework.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q05-1","claimId":"mc-cell-openai-gpt-5-2-codex-q05","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5 Preparedness, PDF p. 8 (printed p. 7)","passage":"We are treating GPT-5.2-Codex as High risk in the Biological and Chemical domain. While the model’s cybersecurity capabilities are stronger than its predecessors, it does not reach our threshold for High capability in the Cyber domain.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q05-2","claimId":"mc-cell-openai-gpt-5-2-codex-q05","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5 Preparedness, PDF p. 8 (printed p. 7)","passage":"It also does not reach High capability in the AI self-improvement domain.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q06-1","claimId":"mc-cell-openai-gpt-5-2-codex-q06","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§1 Introduction, PDF p. 4 (printed p. 3)","passage":"Like other recent models, it is being treated as High capability on biology, and is being deployed with the corresponding suite of safeguards we use for other models in the GPT-5 family.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q07-1","claimId":"mc-cell-openai-gpt-5-2-codex-q07","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.1 Biological and Chemical, PDF p. 9 (printed p. 8)","passage":"As we did for other models in the GPT-5 series, we are treating GPT-5.2-Codex as High risk in the Biological and Chemical domain, and continuing to apply the corresponding safeguards.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q08-1","claimId":"mc-cell-openai-gpt-5-2-codex-q08","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2 Cybersecurity, PDF p. 13 (printed p. 12)","passage":"A review of these results and other relevant information by the Safety Advisory Group and OpenAI leadership determined that GPT-5.2-Codex’s performance—while stronger than earlier models—does not meet the level of consistency needed for High cyber capability.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q09-1","claimId":"mc-cell-openai-gpt-5-2-codex-q09","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.3 AI Self-Improvement, PDF p. 17 (printed p. 16)","passage":"GPT-5.2-Codex performed at a similar capability level to GPT-5.2-Thinking and did not meet our High thresholds.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q09-2","claimId":"mc-cell-openai-gpt-5-2-codex-q09","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.3 AI Self-Improvement, PDF p. 17 (printed p. 16)","passage":"performance in the evaluations below indicate we can rule this out for GPT-5.2-Codex.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q10","claimId":"mc-cell-openai-gpt-5-2-codex-q10","sourceId":"mc-src-gpt-5-2-codex-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q11","claimId":"mc-cell-openai-gpt-5-2-codex-q11","sourceId":"mc-src-gpt-5-2-codex-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q12","claimId":"mc-cell-openai-gpt-5-2-codex-q12","sourceId":"mc-src-gpt-5-2-codex-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q13-1","claimId":"mc-cell-openai-gpt-5-2-codex-q13","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2.4 External Evaluations by Irregular, PDF p. 16 (printed p. 15)","passage":"Irregular, an AI security organization focused on frontier cybersecurity risks, evaluated GPT-5.2-Codex on a subset of a suite of cyberoffensive challenges","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q14","claimId":"mc-cell-openai-gpt-5-2-codex-q14","sourceId":"mc-src-gpt-5-2-codex-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q15-1","claimId":"mc-cell-openai-gpt-5-2-codex-q15","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2 Cybersecurity, PDF p. 13 (printed p. 12)","passage":"A review of these results and other relevant information by the Safety Advisory Group and OpenAI leadership determined that GPT-5.2-Codex’s performance—while stronger than earlier models—does not meet the level of consistency needed for High cyber capability.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q15-2","claimId":"mc-cell-openai-gpt-5-2-codex-q15","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§1 Introduction, PDF p. 4 (printed p. 3)","passage":"Like other recent models, it is being treated as High capability on biology, and is being deployed with the corresponding suite of safeguards we use for other models in the GPT-5 family.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q16-1","claimId":"mc-cell-openai-gpt-5-2-codex-q16","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2 Cybersecurity, PDF p. 12 (printed p. 11)","passage":"Each of these evaluations has important limitations:","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-5-2-codex-q16-2","claimId":"mc-cell-openai-gpt-5-2-codex-q16","sourceId":"mc-src-gpt-5-2-codex-addendum","locator":"§5.1.2 Cybersecurity, PDF p. 13 (printed p. 12)","passage":"Because of these limitations, excelling on all three evaluations is necessary but not sufficient for a model to be considered High in cyber capability.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q01-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q01","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"Cover, PDF p. 1","passage":"Addendum to GPT-6 Luna System Card: October 29, 2026 Update OpenAI 2026-10-29","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q02-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q02","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§1 Introduction, PDF p. 4 (printed p. 3)","passage":"Today, we’re introducing an update to GPT-6 Luna, which we released in ChatGPT on October 7.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q03-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q03","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8 Preparedness, PDF p. 13 (printed p. 12)","passage":"Under our Preparedness Framework, we are treating GPT-6 Luna (October 29) as High capability in both Cybersecurity and Biological and Chemical domains.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q04-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q04","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8 Preparedness, PDF p. 13 (printed p. 12)","passage":"Under our Preparedness Framework, we are treating GPT-6 Luna (October 29) as High capability in both Cybersecurity and Biological and Chemical domains.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q05-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q05","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8 Preparedness, PDF p. 13 (printed p. 12)","passage":"Under our Preparedness Framework, we are treating GPT-6 Luna (October 29) as High capability in both Cybersecurity and Biological and Chemical domains.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q05-2","claimId":"mc-cell-openai-gpt-6-luna-october-29-q05","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8 Preparedness, PDF p. 13 (printed p. 12)","passage":"The model does not reach our High threshold in AI Self-Improvement.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q06-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q06","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§1 Introduction, PDF p. 4 (printed p. 3)","passage":"Based on that assessment, GPT-6 Luna (October 29) uses the same safeguards stack as prior GPT-6 Luna (October 7)","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q06-2","claimId":"mc-cell-openai-gpt-6-luna-october-29-q06","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8 Preparedness, PDF p. 13 (printed p. 12)","passage":"we have implemented the corresponding safeguards described in the GPT-6 Luna (October 7) card.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q07-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q07","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8.1.1 Biological and Chemical Capabilities, PDF p. 13 (printed p. 12)","passage":"We are treating GPT-6 Luna (October 29) as High capability in the Biological and Chemical domain. Below, we report updated results on our High and Critical capability evaluations.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q08-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q08","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8.1.2 Cybersecurity Capabilities, PDF p. 14 (printed p. 13)","passage":"We are treating GPT-6 Luna (October 29) as High capability in the Cybersecurity domain. Below, we report updated results on our High capability evaluations.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q09-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q09","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8.1.3 AI Self-Improvement Capabilities, PDF p. 15 (printed p. 14)","passage":"In AI Self-Improvement, GPT-6 Sol and GPT-6 Luna do not reach our High threshold.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q09-2","claimId":"mc-cell-openai-gpt-6-luna-october-29-q09","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8.1.3, PDF p. 15 (printed p. 14)","passage":"please see the AI Self-Improvement Capabilities section in the GPT-6 Astra card.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q10","claimId":"mc-cell-openai-gpt-6-luna-october-29-q10","sourceId":"mc-src-gpt-6-luna-october-29-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q11-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q11","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§7.0.3 Blocker Deception, Figure 6 caption, PDF p. 12 (printed p. 11)","passage":"Blocker-related disclosure failure rates on coding tasks. Lower is better.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q11-2","claimId":"mc-cell-openai-gpt-6-luna-october-29-q11","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§7.0.2 Respecting Warnings, Figure 5 caption, PDF p. 11 (printed p. 10)","passage":"Conditional failure rates for successful circumvention of environmental restrictions. Lower is better.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q12-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q12","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§4.2 Prompt Injection, PDF p. 7 (printed p. 6)","passage":"we use adversarial training with an automated red-teaming agent to strengthen resistance to malicious instructions","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q12-2","claimId":"mc-cell-openai-gpt-6-luna-october-29-q12","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§4.2 Prompt Injection, PDF p. 7 (printed p. 6)","passage":"For our evaluations, we use the GPT-Red automated evaluations introduced in the GPT-5.6 system card.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q13","claimId":"mc-cell-openai-gpt-6-luna-october-29-q13","sourceId":"mc-src-gpt-6-luna-october-29-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q14","claimId":"mc-cell-openai-gpt-6-luna-october-29-q14","sourceId":"mc-src-gpt-6-luna-october-29-addendum","passageType":"paraphrase"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q15-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q15","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§8.2 Safeguards, PDF p. 15 (printed p. 14)","passage":"The internal report informed our Safety Advisory Group’s recommendation and OpenAI leadership’s determination that these safeguards are sufficient for GPT-6 Luna (October 29) public launch.","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q16-1","claimId":"mc-cell-openai-gpt-6-luna-october-29-q16","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§4.1 Jailbreaks, PDF p. 7 (printed p. 6)","passage":"Our attacker run was cut short due to infrastructure issues","passageType":"quote"},{"id":"mc-ev-cell-openai-gpt-6-luna-october-29-q16-2","claimId":"mc-cell-openai-gpt-6-luna-october-29-q16","sourceId":"mc-src-gpt-6-luna-october-29-addendum","locator":"§2 Model Data and Training, PDF p. 4 (printed p. 3)","passage":"The comparison values for previously launched models that are shown here may reflect later versions of those models, and may vary from the values published at launch.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q01-1","claimId":"mc-cell-xai-grok-4-1-q01","sourceId":"mc-src-grok-4-1-card","locator":"Title block, p. 1","passage":"Grok 4.1 Model Card xAI November 17, 2025","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q02-1","claimId":"mc-cell-xai-grok-4-1-q02","sourceId":"mc-src-grok-4-1-card","locator":"§1 Introduction, p. 1","passage":"It is publicly available through our web and mobile consumer apps.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q03-1","claimId":"mc-cell-xai-grok-4-1-q03","sourceId":"mc-src-grok-4-1-card","locator":"§1 Introduction, p. 1","passage":"In line with our Risk Management Framework (RMF), we measure safety-relevant behaviors across three categories","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q03-2","claimId":"mc-cell-xai-grok-4-1-q03","sourceId":"mc-src-grok-4-1-card","locator":"References, p. 6","passage":"xAI. xai risk management framework, 2025.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q04-1","claimId":"mc-cell-xai-grok-4-1-q04","sourceId":"mc-src-grok-4-1-card","locator":"References, p. 6","passage":"xAI. xai risk management framework, 2025.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q05","claimId":"mc-cell-xai-grok-4-1-q05","sourceId":"mc-src-grok-4-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-1-q06-1","claimId":"mc-cell-xai-grok-4-1-q06","sourceId":"mc-src-grok-4-1-card","locator":"§2.3.2 Results, p. 5","passage":"in response to these results, we have implemented initial mitigations with input filters for restricted chemical and biological knowledge","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q07-1","claimId":"mc-cell-xai-grok-4-1-q07","sourceId":"mc-src-grok-4-1-card","locator":"§2.3.2 Results, p. 4","passage":"Grok 4.1 achieves broadly similar results to Grok 4 and other frontier models on chemical and biological capabilities tests.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q08-1","claimId":"mc-cell-xai-grok-4-1-q08","sourceId":"mc-src-grok-4-1-card","locator":"§2.3.2 Results, p. 5","passage":"Grok 4.1 again performs similarly to other frontier models, but substantially below the level of human cybersecurity experiments","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q09","claimId":"mc-cell-xai-grok-4-1-q09","sourceId":"mc-src-grok-4-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-1-q10-1","claimId":"mc-cell-xai-grok-4-1-q10","sourceId":"mc-src-grok-4-1-card","locator":"§2.3.2 Results, p. 5","passage":"we do not believe it poses risks due to advanced persuasion capabilities","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q11-1","claimId":"mc-cell-xai-grok-4-1-q11","sourceId":"mc-src-grok-4-1-card","locator":"§2.2.1 Deception, p. 3","passage":"Overall, we find that training the model to be honest drops the rate at which Grok 4.1 lies.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q12","claimId":"mc-cell-xai-grok-4-1-q12","sourceId":"mc-src-grok-4-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-1-q13","claimId":"mc-cell-xai-grok-4-1-q13","sourceId":"mc-src-grok-4-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-1-q14","claimId":"mc-cell-xai-grok-4-1-q14","sourceId":"mc-src-grok-4-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-1-q15","claimId":"mc-cell-xai-grok-4-1-q15","sourceId":"mc-src-grok-4-1-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-1-q16-1","claimId":"mc-cell-xai-grok-4-1-q16","sourceId":"mc-src-grok-4-1-card","locator":"§2.1.2 Evaluations, p. 2","passage":"results in previous model cards were reported with an error in the evaluation settings, where only the English prompts were evaluated","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-1-q16-2","claimId":"mc-cell-xai-grok-4-1-q16","sourceId":"mc-src-grok-4-1-card","locator":"§2.3.2 Results, pp. 4-5","passage":"the human baselines collected in previous work likely underestimate the … performance of high-context experts","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q01-1","claimId":"mc-cell-xai-grok-4-7-q01","sourceId":"mc-src-grok-4-7-card","locator":"Cover, PDF p. 1","passage":"Model Card: Grok 4.7 September 21, 2026 Revision: 2026-09-21","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q02-1","claimId":"mc-cell-xai-grok-4-7-q02","sourceId":"mc-src-grok-4-7-card","locator":"§1.1 Overview, PDF p. 5 (printed p. 4)","passage":"It is available for use through the following channels","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q02-2","claimId":"mc-cell-xai-grok-4-7-q02","sourceId":"mc-src-grok-4-7-card","locator":"§1.1 Overview, PDF p. 5 (printed p. 4)","passage":"SpaceXAI plans to add Grok 4.7 to its consumer surfaces (web, mobile apps, and Grok-in-X on the X platform) at a later date.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q03-1","claimId":"mc-cell-xai-grok-4-7-q03","sourceId":"mc-src-grok-4-7-card","locator":"§7 Biological and chemical capabilities and safeguards, PDF p. 20 (printed p. 19)","passage":"treated as safety-threshold evaluations under the xAI Frontier Artificial Intelligence Framework (FAIF)","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q03-2","claimId":"mc-cell-xai-grok-4-7-q03","sourceId":"mc-src-grok-4-7-card","locator":"References, item 16, PDF p. 29 (printed p. 28)","passage":"xAI Frontier Artificial Intelligence Framework (FAIF). xAI, June 30, 2026.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q04-1","claimId":"mc-cell-xai-grok-4-7-q04","sourceId":"mc-src-grok-4-7-card","locator":"References, item 16, PDF p. 29 (printed p. 28)","passage":"xAI Frontier Artificial Intelligence Framework (FAIF). xAI, June 30, 2026.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q05-1","claimId":"mc-cell-xai-grok-4-7-q05","sourceId":"mc-src-grok-4-7-card","locator":"§7 Biological and chemical capabilities and safeguards, PDF p. 20 (printed p. 19)","passage":"Grok 4.7 scores below the FAIF safety thresholds on dual-use knowledge","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q06-1","claimId":"mc-cell-xai-grok-4-7-q06","sourceId":"mc-src-grok-4-7-card","locator":"§9.3 CBRN / weapons refusals, PDF p. 25 (printed p. 24)","passage":"dedicated input and topical filters target biological and chemical weapons-related abuse","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q07-1","claimId":"mc-cell-xai-grok-4-7-q07","sourceId":"mc-src-grok-4-7-card","locator":"§7 Biological and chemical capabilities and safeguards, PDF p. 20 (printed p. 19)","passage":"Grok 4.7 scores below the FAIF safety thresholds on dual-use knowledge","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q08-1","claimId":"mc-cell-xai-grok-4-7-q08","sourceId":"mc-src-grok-4-7-card","locator":"§6 Cyber capabilities and safeguards, PDF p. 17 (printed p. 16)","passage":"Grok 4.7 shows small cybersecurity-relevant capability gains over Grok 4.6","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q09","claimId":"mc-cell-xai-grok-4-7-q09","sourceId":"mc-src-grok-4-7-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-7-q10","claimId":"mc-cell-xai-grok-4-7-q10","sourceId":"mc-src-grok-4-7-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-7-q11-1","claimId":"mc-cell-xai-grok-4-7-q11","sourceId":"mc-src-grok-4-7-card","locator":"§11.1 MASK-Rectified, PDF p. 26 (printed p. 25)","passage":"we test whether Grok 4.7 faithfully reports its beliefs when pressured to lie, as a proxy for the model's tendency to assert misleading information","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q12","claimId":"mc-cell-xai-grok-4-7-q12","sourceId":"mc-src-grok-4-7-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-7-q13-1","claimId":"mc-cell-xai-grok-4-7-q13","sourceId":"mc-src-grok-4-7-card","locator":"§2.2 DeepSWE v1.1, note, PDF p. 8 (printed p. 7)","passage":"Results reported are taken from evaluations conducted by Datacurve.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q13-2","claimId":"mc-cell-xai-grok-4-7-q13","sourceId":"mc-src-grok-4-7-card","locator":"§6.4 CathedralBench, PDF p. 19 (printed p. 18)","passage":"CathedralBench is an independent, third party evaluation of cyber performance.","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q13-3","claimId":"mc-cell-xai-grok-4-7-q13","sourceId":"mc-src-grok-4-7-card","locator":"§6 Cyber capabilities and safeguards, PDF p. 17 (printed p. 16)","passage":"We additionally provided an unrestricted configuration of Grok 4.7 to third-party evaluators","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q14","claimId":"mc-cell-xai-grok-4-7-q14","sourceId":"mc-src-grok-4-7-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-7-q15","claimId":"mc-cell-xai-grok-4-7-q15","sourceId":"mc-src-grok-4-7-card","passageType":"paraphrase"},{"id":"mc-ev-cell-xai-grok-4-7-q16-1","claimId":"mc-cell-xai-grok-4-7-q16","sourceId":"mc-src-grok-4-7-card","locator":"§7.1 BioSecBench, PDF p. 20 (printed p. 19)","passage":"the restricted task sets characterize the tested checkpoint and harness rather than establishing general biological safety or reliability","passageType":"quote"},{"id":"mc-ev-cell-xai-grok-4-7-q16-2","claimId":"mc-cell-xai-grok-4-7-q16","sourceId":"mc-src-grok-4-7-card","locator":"§2.3 Terminal-Bench 4.0, PDF p. 9 (printed p. 8)","passage":"absolute scores remain sensitive to the agent harness","passageType":"quote"}],"at":{"entities":[21,22,26,42,43,44,45,46,47,48,49,50,51,52,53,54,56,57,114,2133,2134,2135,2136,2137,2138,2139,2140,2141,2142,2143,2144,2145,2146,2147,2148,2149,2150],"claims":[3388,3390,5834,5836,5840,5850,5852,5854,5857,5858,5859,5860,5861,5862,5863,5864,5865,5866,5867,5868,5869,5870,5871,5872,5873,5874,5875,5876,5877,5878,5879,5880,5881,5882,5883,5884,5885,5886,5887,5888,5889,5890,5891,5892,5893,5894,5895,5896,5897,5898,5899,5900,5901,5902,5903,5904,5905,5906,5907,5908,5909,5910,5911,5912,5913,5914,5915,5916,5917,5918,5919,5920,5921,5922,5923,5924,5925,5926,5927,5928,5929,5930,5931,5932,5933,5934,5935,5936,5937,5938,5939,5940,5941,5942,5943,5944,5945,5946,5947,5948,5949,5950,5951,5952,5953,5954,5955,5956,5957,5958,5959,5960,5961,5962,5963,5964,5965,5966,5967,5968,5969,5970,5971,5972,5973,5974,5975,5976,5977,5978,5979,5980,5981,5982,5983,5984,5985,5986,5987,5988,5989,5990,5991,5992,5993,5994,5995,5996,5997,5998,5999,6000,6001,6002,6003,6004,6005,6006,6007,6008,6009,6010,6011,6012,6013,6014,6015,6016,6017,6018,6019,6020,6021,6022,6023,6024,6025,6026,6027,6028,6029,6030,6031,6032,6033,6034,6035,6036,6037,6038,6039,6040,6041,6042,6043,6044,6045,6046,6047,6048,6049,6050,6051,6052,6053,6054,6055,6056,6057,6058,6059,6060,6061,6062,6063,6064,6065,6066,6067,6068,6069,6070,6071,6072,6073,6074,6075,6076,6077,6078,6079,6080,6081,6082,6083,6084,6085,6086,6087,6088,6089,6090,6091,6092,6093,6094,6095,6096,6097,6098,6099,6100,6101,6102,6103,6104,6105,6106,6107,6108,6109,6110,6111,6112,6113,6114,6115,6116,6117,6118,6119,6120,6121,6122,6123,6124,6125,6126,6127,6128,6129,6130,6131,6132,6133,6134,6135,6136,6137,6138,6139,6140,6141,6142,6143,6144,6145,6146,6147,6148,6149,6150,6151,6152,6153,6154,6155,6156,6157,6158,6159,6160,6161,6162],"sources":[13,116,2014,2015,2016,2017,2018,2019,2020,2021,2022,2023,2024,2025,2026,2027,2029,2030,2031,2032],"evidence":[7983,7984,7985,7986,7987,7988,7989,7990,7991,7992,7993,7994,7995,7996,7997,7998,7999,8000,8001,8002,8003,8004,8005,8006,8007,8008,8009,8010,8011,8012,8013,8014,8015,8016,8017,8018,8019,8020,8021,8022,8023,8024,8025,8026,8027,8028,8029,8030,8031,8032,8033,8034,8035,8036,8037,8038,8039,8040,8041,8042,8043,8044,8045,8046,8047,8048,8049,8050,8051,8052,8053,8054,8055,8056,8057,8058,8059,8060,8061,8062,8063,8064,8065,8066,8067,8068,8069,8070,8071,8072,8073,8074,8075,8076,8077,8078,8079,8080,8081,8082,8083,8084,8085,8086,8087,8088,8089,8090,8091,8092,8093,8094,8095,8096,8097,8098,8099,8100,8101,8102,8103,8104,8105,8106,8107,8108,8109,8110,8111,8112,8113,8114,8115,8116,8117,8118,8119,8120,8121,8122,8123,8124,8125,8126,8127,8128,8129,8130,8131,8132,8133,8134,8135,8136,8137,8138,8139,8140,8141,8142,8143,8144,8145,8146,8147,8148,8149,8150,8151,8152,8153,8154,8155,8156,8157,8158,8159,8160,8161,8162,8163,8164,8165,8166,8167,8168,8169,8170,8171,8172,8173,8174,8175,8176,8177,8178,8179,8180,8181,8182,8183,8184,8185,8186,8187,8188,8189,8190,8191,8192,8193,8194,8195,8196,8197,8198,8199,8200,8201,8202,8203,8204,8205,8206,8207,8208,8209,8210,8211,8212,8213,8214,8215,8216,8217,8218,8219,8220,8221,8222,8223,8224,8225,8226,8227,8228,8229,8230,8231,8232,8233,8234,8235,8236,8237,8238,8239,8240,8241,8242,8243,8244,8245,8246,8247,8248,8249,8250,8251,8252,8253,8254,8255,8256,8257,8258,8259,8260,8261,8262,8263,8264,8265,8266,8267,8268,8269,8270,8271,8272,8273,8274,8275,8276,8277,8278,8279,8280,8281,8282,8283,8284,8285,8286,8287,8288,8289,8290,8291,8292,8293,8294,8295,8296,8297,8298,8299,8300,8301,8302,8303,8304,8305,8306,8307,8308,8309,8310,8311,8312,8313,8314,8315,8316,8317,8318,8319,8320,8321,8322,8323,8324,8325,8326,8327,8328,8329,8330,8331,8332,8333,8334,8335,8336,8337,8338,8339,8340,8341,8342,8343,8344,8345,8346,8347,8348,8349,8350,8351,8352,8353,8354,8355,8356,8357,8358,8359,8360,8361,8362,8363,8364,8365,8366,8367,8368,8369,8370,8371,8372,8373,8374,8375,8376,8377,8378,8379,8380,8381,8382,8383,8384,8385,8386,8387,8388,8389,8390,8391,8392]},"index":{"byClaim":[["mc-cell-anthropic-claude-2-1-q01",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q02",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q03",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q04",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q05",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q06",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q07",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q08",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q09",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q10",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q11",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q12",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q13",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q14",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q15",["compare:model-cards"]],["mc-cell-anthropic-claude-2-1-q16",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q01",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q02",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q03",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q04",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q05",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q06",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q07",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q08",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q09",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q10",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q11",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q12",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q13",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q14",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q15",["compare:model-cards"]],["mc-cell-anthropic-claude-3-5-haiku-sonnet-new-q16",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q01",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q02",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q03",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q04",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q05",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q06",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q07",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q08",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q09",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q10",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q11",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q12",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q13",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q14",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q15",["compare:model-cards"]],["mc-cell-anthropic-claude-opus-4-5-q16",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q01",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q02",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q03",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q04",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q05",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q06",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q07",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q08",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q09",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q10",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q11",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q12",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q13",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q14",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q15",["compare:model-cards"]],["mc-cell-anthropic-claude-haiku-5-5-q16",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q01",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q02",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q03",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q04",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q05",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q06",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q07",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q08",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q09",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q10",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q11",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q12",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q13",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q14",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q15",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-3-flash-q16",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q01",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q02",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q03",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q04",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q05",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q06",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q07",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q08",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q09",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q10",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q11",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q12",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q13",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q14",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q15",["compare:model-cards"]],["mc-cell-google-deepmind-gemini-4-argon-q16",["compare:model-cards"]],["mc-cell-meta-llama-2-q01",["compare:model-cards"]],["mc-cell-meta-llama-2-q02",["compare:model-cards"]],["mc-cell-meta-llama-2-q06",["compare:model-cards"]],["mc-cell-meta-llama-2-q07",["compare:model-cards"]],["mc-cell-meta-llama-2-q08",["compare:model-cards"]],["mc-cell-meta-llama-2-q09",["compare:model-cards"]],["mc-cell-meta-llama-2-q10",["compare:model-cards"]],["mc-cell-meta-llama-2-q11",["compare:model-cards"]],["mc-cell-meta-llama-2-q12",["compare:model-cards"]],["mc-cell-meta-llama-2-q13",["compare:model-cards"]],["mc-cell-meta-llama-2-q14",["compare:model-cards"]],["mc-cell-meta-llama-2-q15",["compare:model-cards"]],["mc-cell-meta-llama-2-q16",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q01",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q02",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q06",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q07",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q08",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q09",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q10",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q11",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q12",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q13",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q14",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q15",["compare:model-cards"]],["mc-cell-meta-llama-3-3-q16",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q01",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q02",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q03",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q04",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q05",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q06",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q07",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q08",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q09",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q10",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q11",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q12",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q13",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q14",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q15",["compare:model-cards"]],["mc-cell-meta-cwm-preparedness-q16",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q01",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q02",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q03",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q04",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q05",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q06",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q07",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q08",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q09",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q10",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q11",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q12",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q13",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q14",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q15",["compare:model-cards"]],["mc-cell-meta-muse-spark-1-3-q16",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q01",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q02",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q06",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q07",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q08",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q09",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q10",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q11",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q12",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q13",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q14",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q15",["compare:model-cards"]],["mc-cell-microsoft-phi-2-q16",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q01",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q02",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q06",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q07",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q08",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q09",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q10",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q11",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q12",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q13",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q14",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q15",["compare:model-cards"]],["mc-cell-microsoft-phi-4-q16",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q01",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q02",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q03",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q04",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q05",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q06",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q07",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q08",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q09",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q10",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q11",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q12",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q13",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q14",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q15",["compare:model-cards"]],["mc-cell-microsoft-phi-4-mini-flash-reasoning-q16",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q01",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q02",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q03",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q04",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q05",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q06",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q07",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q08",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q09",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q10",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q11",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q12",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q13",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q14",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q15",["compare:model-cards"]],["mc-cell-microsoft-mai-cyber-1-1-flash-q16",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q01",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q02",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q06",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q07",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q08",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q09",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q10",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q11",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q12",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q13",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q14",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q15",["compare:model-cards"]],["mc-cell-openai-gpt-4v-q16",["compare:model-cards"]],["mc-cell-openai-o1-q01",["compare:model-cards"]],["mc-cell-openai-o1-q02",["compare:model-cards"]],["mc-cell-openai-o1-q03",["compare:model-cards"]],["mc-cell-openai-o1-q04",["compare:model-cards"]],["mc-cell-openai-o1-q05",["compare:model-cards"]],["mc-cell-openai-o1-q06",["compare:model-cards"]],["mc-cell-openai-o1-q07",["compare:model-cards"]],["mc-cell-openai-o1-q08",["compare:model-cards"]],["mc-cell-openai-o1-q09",["compare:model-cards"]],["mc-cell-openai-o1-q10",["compare:model-cards"]],["mc-cell-openai-o1-q11",["compare:model-cards"]],["mc-cell-openai-o1-q12",["compare:model-cards"]],["mc-cell-openai-o1-q13",["compare:model-cards"]],["mc-cell-openai-o1-q14",["compare:model-cards"]],["mc-cell-openai-o1-q15",["compare:model-cards"]],["mc-cell-openai-o1-q16",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q01",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q02",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q03",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q04",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q05",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q06",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q07",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q08",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q09",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q10",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q11",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q12",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q13",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q14",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q15",["compare:model-cards"]],["mc-cell-openai-gpt-5-2-codex-q16",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q01",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q02",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q03",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q04",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q05",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q06",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q07",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q08",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q09",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q10",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q11",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q12",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q13",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q14",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q15",["compare:model-cards"]],["mc-cell-openai-gpt-6-luna-october-29-q16",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q01",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q02",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q03",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q04",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q05",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q06",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q07",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q08",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q09",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q10",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q11",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q12",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q13",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q14",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q15",["compare:model-cards"]],["mc-cell-xai-grok-4-1-q16",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q01",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q02",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q03",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q04",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q05",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q06",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q07",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q08",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q09",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q10",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q11",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q12",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q13",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q14",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q15",["compare:model-cards"]],["mc-cell-xai-grok-4-7-q16",["compare:model-cards"]]],"marks":[],"chronicle":{"years":[{"year":1637,"topics":{"ideas":2},"stated":0},{"year":1843,"topics":{"ideas":1},"stated":0},{"year":1863,"topics":{"ideas":10},"stated":0},{"year":1865,"topics":{"ideas":2},"stated":0},{"year":1872,"topics":{"ideas":12},"stated":0},{"year":1923,"topics":{"ideas":3},"stated":0},{"year":1943,"topics":{"ideas":1},"stated":0},{"year":1947,"topics":{"ideas":4},"stated":0},{"year":1948,"topics":{"ideas":7},"stated":0},{"year":1949,"topics":{"ideas":4},"stated":0},{"year":1950,"topics":{"ideas":11},"stated":0},{"year":1951,"topics":{"ideas":14},"stated":0},{"year":1952,"topics":{"ideas":2},"stated":0},{"year":1955,"topics":{"ideas":31},"stated":0},{"year":1956,"topics":{"ideas":1},"stated":0},{"year":1957,"topics":{"ideas":1},"stated":0},{"year":1958,"topics":{"ideas":3},"stated":0},{"year":1959,"topics":{"ideas":7},"stated":0},{"year":1960,"topics":{"ideas":12},"stated":0},{"year":1961,"topics":{"ideas":14},"stated":0},{"year":1965,"topics":{"ideas":8},"stated":0},{"year":1966,"topics":{"ideas":21},"stated":0},{"year":1968,"topics":{"ideas":5},"stated":0},{"year":1969,"topics":{"ideas":2},"stated":0},{"year":1971,"topics":{"ideas":4},"stated":0},{"year":1972,"topics":{"ideas":1},"stated":0},{"year":1973,"topics":{"ideas":2},"stated":0},{"year":1974,"topics":{"ideas":2},"stated":0},{"year":1975,"topics":{"ideas":1},"stated":0},{"year":1976,"topics":{"ideas":1},"stated":0},{"year":1977,"topics":{"people":2,"ideas":8},"stated":0},{"year":1979,"topics":{"ideas":10},"stated":0},{"year":1980,"topics":{"ideas":2},"stated":0},{"year":1981,"topics":{"people":2},"stated":0},{"year":1982,"topics":{"ideas":4},"stated":0},{"year":1983,"topics":{"ideas":5},"stated":0},{"year":1984,"topics":{"ideas":21},"stated":0},{"year":1985,"topics":{"ideas":9},"stated":0},{"year":1986,"topics":{"ideas":18},"stated":0},{"year":1987,"topics":{"ideas":1},"stated":0},{"year":1988,"topics":{"ideas":6},"stated":0},{"year":1989,"topics":{"ideas":25},"stated":0},{"year":1990,"topics":{"ideas":9},"stated":0},{"year":1991,"topics":{"ideas":14},"stated":0},{"year":1992,"topics":{"ideas":15},"stated":0},{"year":1993,"topics":{"ideas":24,"people":3},"stated":0},{"year":1994,"topics":{"ideas":18},"stated":0},{"year":1995,"topics":{"ideas":10,"people":1},"stated":0},{"year":1996,"topics":{"ideas":83},"stated":0},{"year":1997,"topics":{"ideas":19},"stated":0},{"year":1998,"topics":{"ideas":53},"stated":0},{"year":1999,"topics":{"ideas":32,"people":1,"institutions":1},"stated":0},{"year":2000,"topics":{"ideas":76},"stated":0},{"year":2001,"topics":{"ideas":53},"stated":0},{"year":2002,"topics":{"ideas":31},"stated":0},{"year":2003,"topics":{"ideas":9},"stated":0},{"year":2004,"topics":{"ideas":14,"people":3},"stated":0},{"year":2005,"topics":{"ideas":20,"people":1},"stated":0},{"year":2006,"topics":{"ideas":26,"people":1},"stated":0},{"year":2007,"topics":{"ideas":21},"stated":0},{"year":2008,"topics":{"ideas":67,"people":3,"institutions":1},"stated":0},{"year":2009,"topics":{"ideas":12,"people":1},"stated":0},{"year":2010,"topics":{"people":3,"ideas":13,"institutions":1,"governance":4},"stated":0},{"year":2011,"topics":{"people":3,"ideas":16,"institutions":2},"stated":0},{"year":2012,"topics":{"people":4,"ideas":20,"institutions":2},"stated":0},{"year":2013,"topics":{"ideas":47,"people":4,"institutions":2,"capital":1},"stated":0},{"year":2014,"topics":{"institutions":2,"ideas":60,"people":5},"stated":0},{"year":2015,"topics":{"people":23,"ideas":133,"institutions":3,"capital":1},"stated":0},{"year":2016,"topics":{"ideas":88,"people":2},"stated":0},{"year":2017,"topics":{"people":16,"institutions":2,"ideas":130,"governance":2},"stated":0},{"year":2018,"topics":{"people":9,"ideas":159,"institutions":8,"governance":1},"stated":0},{"year":2019,"topics":{"capital":1,"compute":1,"people":11,"governance":19,"ideas":176,"institutions":4},"stated":0},{"year":2020,"topics":{"people":6,"ideas":98,"institutions":1,"governance":2},"stated":0},{"year":2021,"topics":{"people":27,"ideas":177,"compute":1,"institutions":23,"governance":4,"capital":6},"stated":0},{"year":2022,"topics":{"people":15,"ideas":228,"institutions":6,"governance":3,"compute":2,"capital":1},"stated":0},{"year":2023,"topics":{"capital":14,"compute":7,"institutions":25,"people":55,"governance":211,"ideas":411},"stated":0},{"year":2024,"topics":{"capital":31,"compute":19,"governance":299,"people":103,"ideas":180,"institutions":32},"stated":1},{"year":2025,"topics":{"governance":635,"compute":69,"capital":61,"people":145,"institutions":26,"ideas":180},"stated":7},{"year":2026,"topics":{"governance":533,"capital":30,"compute":32,"institutions":23,"ideas":97,"people":139},"stated":19},{"year":2027,"topics":{},"stated":8},{"year":2028,"topics":{},"stated":6},{"year":2029,"topics":{},"stated":3},{"year":2030,"topics":{},"stated":3},{"year":2032,"topics":{},"stated":2},{"year":2035,"topics":{},"stated":2},{"year":2039,"topics":{},"stated":1},{"year":2042,"topics":{},"stated":1},{"year":2050,"topics":{},"stated":1}],"scheduled":[],"statedBy":[]},"collections":[{"target":{"kind":"propositions","id":"antecedent-texts"},"title":"Antecedent texts: machine intelligence before 1965","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"superintelligence"},"title":"Propositions about superintelligence, 1863–2025","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"risk-debate"},"title":"Arguments about catastrophic risk from advanced AI, 1863–2026","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"yudkowsky-arguments"},"title":"Eliezer Yudkowsky’s arguments on common claims about AI, 1996–2026","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"machine-succession"},"title":"Machine succession, scale and acceleration: propositions in located texts, 1863–2026","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"economics-of-machine-intelligence"},"title":"Economics of machine intelligence: texts from 1952 to 2026 that state, contest or bear on its propositions","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"orbital-compute-arguments"},"title":"Arguments about compute in orbit","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"ai-bio-arguments"},"title":"Arguments about AI and biological risk","claimIds":[],"records":[]},{"target":{"kind":"propositions","id":"ai-in-science"},"title":"AI in science","claimIds":[],"records":[]},{"target":{"kind":"documents"},"title":"Frontier safety frameworks","claimIds":[]},{"target":{"kind":"compare","id":"frontier-safety-frameworks"},"title":"Compare frameworks","claimIds":[]},{"target":{"kind":"control","id":"openai"},"title":"OpenAI’s control structure, 2015–2026","claimIds":[]},{"target":{"kind":"control","id":"anthropic"},"title":"Anthropic’s control structure, 2021–2026","claimIds":[]},{"target":{"kind":"commitments"},"title":"External commitments and obligations","claimIds":[]},{"target":{"kind":"question","id":"frameworks-to-obligations"},"title":"Frameworks, commitments and legal obligations, 2023–2026","claimIds":[]},{"target":{"kind":"reading"},"title":"Founding and early leadership","claimIds":[]}]},"version":"1tzkjxk"}